The development of a deep learning model for automated segmentation of the robotic pancreaticojejunostomy

BACKGROUND

Minimally invasive surgery provides an unprecedented opportunity to review video for assessing surgical performance. Surgical video analysis is time-consuming and expensive. Deep learning provides an alternative for analysis. Robotic pancreaticoduodenectomy (RPD) is a complex and morbid operation. Surgeon technical performance of pancreaticojejunostomy (PJ) has been associated with postoperative pancreatic fistula. In this work, we aimed to utilize deep learning to automatically segment PJ RPD videos.

METHODS

This was a retrospective review of prospectively collected videos from 2011 to 2022 that were in libraries at tertiary referral centers, including 111 PJ videos. Each frame of a robotic PJ video was categorized based on 6 tasks. A 3D convolutional neural network was trained for frame-level visual feature extraction and classification. All the videos were manually annotated for the start and end of each task.

RESULTS

Of the 100 videos assessed, 60 videos were used for the training the model, 10 for hyperparameter optimization, and 30 for the testing of performance. All the frames were extracted (6 frames/second) and annotated. The accuracy and mean per-class F1 scores were 88.01% and 85.34% for tasks.

CONCLUSION

The deep learning model performed well for automated segmentation of PJ videos. Future work will focus on skills assessment and outcome prediction.

Use of MRI-based deep learning radiomics to diagnose sacroiliitis related to axial spondyloarthritis

OBJECTIVES

This study aimed to evaluate the performance of a deep learning radiomics (DLR) model, which integrates multimodal MRI features and clinical information, in diagnosing sacroiliitis related to axial spondyloarthritis (axSpA).

MATERIAL & METHODS

A total of 485 patients diagnosed with sacroiliitis related to axSpA (n = 288) or non-sacroiliitis (n = 197) by sacroiliac joint (SIJ) MRI between May 2018 and October 2022 were retrospectively included in this study. The patients were randomly divided into training (n = 388) and testing (n = 97) cohorts. Data were collected using three MRI scanners. We applied a convolutional neural network (CNN) called 3D U-Net for automated SIJ segmentation. Additionally, three CNNs (ResNet50, ResNet101, and DenseNet121) were used to diagnose axSpA-related sacroiliitis using a single modality. The prediction results of all the CNN models across different modalities were integrated using a stacking method based on different algorithms to construct ensemble models, and the optimal ensemble model was used as DLR signature. A combined model incorporating DLR signature with clinical factors was developed using multivariable logistic regression. The performance of the models was evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA).

RESULTS

Automated deep learning-based segmentation and manual delineation showed good correlation. ResNet50, as the optimal basic model, achieved an area under the curve (AUC) and accuracy of 0.839 and 0.804, respectively. The combined model yielded the highest performance in diagnosing axSpA-related sacroiliitis (AUC: 0.910; accuracy: 0.856) and outperformed the best ensemble model (AUC: 0.868; accuracy: 0.825) (all P < 0.05). Moreover, the DCA showed good clinical utility in the combined model.

CONCLUSION

We developed a diagnostic model for axSpA-related sacroiliitis by combining the DLR signature with clinical factors, which resulted in excellent diagnostic performance.

Craniofacial phenotyping with fetal MRI: a feasibility study of 3D visualisation, segmentation, surface-rendered and physical models

This study explores the potential of 3D Slice-to-Volume Registration (SVR) motion-corrected fetal MRI for craniofacial assessment, traditionally used only for fetal brain analysis. In addition, we present the first description of an automated pipeline based on 3D Attention UNet trained for 3D fetal MRI craniofacial segmentation, followed by surface refinement. Results of 3D printing of selected models are also presented.Qualitative analysis of multiplanar volumes, based on the SVR output and surface segmentations outputs, were assessed with computer and printed models, using standardised protocols that we developed for evaluating image quality and visibility of diagnostic craniofacial features. A test set of 25, postnatally confirmed, Trisomy 21 fetal cases (24-36 weeks gestational age), revealed that 3D reconstructed T2 SVR images provided 66-100% visibility of relevant craniofacial and head structures in the SVR output, and 20-100% and 60-90% anatomical visibility was seen for the baseline and refined 3D computer surface model outputs respectively. Furthermore, 12 of 25 cases, 48%, of refined surface models demonstrated good or excellent overall quality with a further 9 cases, 36%, demonstrating moderate quality to include facial, scalp and external ears. Additional 3D printing of 12 physical real-size models (20-36 weeks gestational age) revealed good/excellent overall quality in all cases and distinguishable features between healthy control cases and cases with confirmed anomalies, with only minor manual adjustments required before 3D printing.Despite varying image quality and data heterogeneity, 3D T2w SVR reconstructions and models provided sufficient resolution for the subjective characterisation of subtle craniofacial features. We also contributed a publicly accessible online 3D T2w MRI atlas of the fetal head, validated for accurate representation of normal fetal anatomy.Future research will focus on quantitative analysis, optimizing the pipeline, and exploring diagnostic, counselling, and educational applications in fetal craniofacial assessment.

Vascular compactness of unruptured brain arteriovenous malformation predicts risk of hemorrhage after stereotactic radiosurgery

The aim of the study was to investigate whether morphology (i.e. compact/diffuse) of brain arteriovenous malformations (bAVMs) correlates with the incidence of hemorrhagic events in patients receiving Stereotactic Radiosurgery (SRS) for unruptured bAVMs. This retrospective study included 262 adult patients with unruptured bAVMs who underwent upfront SRS. Hemorrhagic events were defined as evidence of blood on CT or MRI. The morphology of bAVMs was evaluated using automated segmentation which calculated the proportion of vessel, brain tissue, and cerebrospinal fluid in bAVMs on T2-weighted MRI. Compactness index, defined as the ratio of vessel to brain tissue, categorized bAVMs into compact and diffuse types based on the optimal cutoff. Cox proportional hazard model was used to identify the independent factors for post-SRS hemorrhage. The median clinical follow-ups was 62.1 months. Post-SRS hemorrhage occurred in 13 (5.0%) patients and one of them had two bleeds, resulting in an annual bleeding rate of 0.8%. Multivariable analysis revealed bAVM morphology (compact versus diffuse), bAVM volume, and prescribed margin dose were significant predictors. The post-SRS hemorrhage rate increased with larger bAVM volume only among the diffuse nidi (1.7 versus 14.9 versus 30.6 hemorrhage per 1000 person-years in bAVM volume < 20 cm3 versus 20-40 cm3 versus > 40 cm3; p = 0.022). The significantly higher post-SRS hemorrhage rate of Spetzler-Martin grade IV-V compared with grade I-III bAVMs (20.0 versus 3.3 hemorrhages per 1000 person-years; p = 0.001) mainly originated from the diffuse bAVMs rather than the compact subgroup (30.9 versus 4.8 hemorrhages per 1000 person-years; p = 0.035). Compact and smaller bAVMs, with higher prescribed margin dose harbor lower risks of post-SRS hemorrhage. The post-SRS hemorrhage rate exceeded 2.2% annually within the diffuse and large (> 40 cm3) bAVMs and the diffuse Spetzler-Martin IV-V bAVMs. These findings may help guide patient selection of SRS for the unruptured bAVMs.

Automated abdominal adipose tissue segmentation and volume quantification on longitudinal MRI using 3D convolutional neural networks with multi-contrast inputs

OBJECTIVE

Increased subcutaneous and visceral adipose tissue (SAT/VAT) volume is associated with risk for cardiometabolic diseases. This work aimed to develop and evaluate automated abdominal SAT/VAT segmentation on longitudinal MRI in adults with overweight/obesity using attention-based competitive dense (ACD) 3D U-Net and 3D nnU-Net with full field-of-view volumetric multi-contrast inputs.

MATERIALS AND METHODS

920 adults with overweight/obesity were scanned twice at multiple 3 T MRI scanners and institutions. The first scan was divided into training/validation/testing sets (n = 646/92/182). The second scan from the subjects in the testing set was used to evaluate the generalizability for longitudinal analysis. Segmentation performance was assessed by measuring Dice scores (DICE-SAT, DICE-VAT), false negatives (FN), and false positives (FP). Volume agreement was assessed using the intraclass correlation coefficient (ICC).

RESULTS

ACD 3D U-Net achieved rapid (< 4.8 s/subject) segmentation with high DICE-SAT (median ≥ 0.994) and DICE-VAT (median ≥ 0.976), small FN (median ≤ 0.7%), and FP (median ≤ 1.1%). 3D nnU-Net yielded rapid (< 2.5 s/subject) segmentation with similar DICE-SAT (median ≥ 0.992), DICE-VAT (median ≥ 0.979), FN (median ≤ 1.1%) and FP (median ≤ 1.2%). Both models yielded excellent agreement in SAT/VAT volume versus reference measurements (ICC > 0.997) in longitudinal analysis.

DISCUSSION

ACD 3D U-Net and 3D nnU-Net can be automated tools to quantify abdominal SAT/VAT volume rapidly, accurately, and longitudinally in adults with overweight/obesity.

Automatic lung segmentation of magnetic resonance images: A new approach applied to healthy volunteers undergoing enhanced Deep-Inspiration-Breath-Hold for motion-mitigated 4D proton therapy of lung tumors

Background and purpose

Respiratory suppression techniques represent an effective motion mitigation strategy for 4D-irradiation of lung tumors with protons. A magnetic resonance imaging (MRI)-based study applied and analyzed methods for this purpose, including enhanced Deep-Inspiration-Breath-Hold (eDIBH). Twenty-one healthy volunteers (41-58 years) underwent thoracic MR scans in four imaging sessions containing two eDIBH-guided MRIs per session to simulate motion-dependent irradiation conditions. The automated MRI segmentation algorithm presented here was critical in determining the lung volumes (LVs) achieved during eDIBH.

Materials and methods

The study included 168 MRIs acquired under eDIBH conditions. The lung segmentation algorithm consisted of four analysis steps: (i) image preprocessing, (ii) MRI histogram analysis with thresholding, (iii) automatic segmentation, (iv) 3D-clustering. To validate the algorithm, 46 eDIBH-MRIs were manually contoured. Sørensen-Dice similarity coefficients (DSCs) and relative deviations of LVs were determined as similarity measures. Assessment of intrasessional and intersessional LV variations and their differences provided estimates of statistical and systematic errors.

Results

Lung segmentation time for 100 2D-MRI planes was ∼ 10 s. Compared to manual lung contouring, the median DSC was 0.94 with a lower 95 % confidence level (CL) of 0.92. The relative volume deviations yielded a median value of 0.059 and 95 % CLs of -0.013 and 0.13. Artifact-based volume errors, mainly of the trachea, were estimated. Estimated statistical and systematic errors ranged between 6 and 8 %.

Conclusions

The presented analytical algorithm is fast, precise, and readily available. The results are comparable to time-consuming, manual segmentations and other automatic segmentation approaches. Post-processing to remove image artifacts is under development.

Electron Microscopy Techniques for 3D Plant ER Imaging

The endoplasmic reticulum (ER) forms an extensive network in plant cells. In leaf cells and vacuolated root cells it is mainly restricted to the cortex, whereas in the root meristem the cortical and cytoplasmic ER takes up a large volume throughout the entire cell. Only 3D electron microscopy provides sufficient resolution to understand the spatial organization of the ER in the root. Here we present two protocols for 3D EM imaging of the ER across a range of scales. For large-scale ER structure analysis, we describe selective ER staining with ZIO that allows for automated or semi-automated ER segmentation. For smaller regions of ER, we describe high-pressure freezing, which enables almost instantaneous fixation of plant tissues but without organelle specific staining. These fixation and staining techniques are suitable for a range of imaging modalities, including serial sections, array tomography, serial block face-scanning electron microscopy (SBF-SEM), or focused ion beam (FIB) SEM.

An overview of clinical machine learning applications in neurology

Machine learning techniques for clinical applications are evolving, and the potential impact this will have on clinical neurology is important to recognize. By providing a broad overview on this growing paradigm of clinical tools, this article aims to help healthcare professionals in neurology prepare to navigate both the opportunities and challenges brought on through continued advancements in machine learning. This narrative review first elaborates on how machine learning models are organized and implemented. Machine learning tools are then classified by clinical application, with examples of uses within neurology described in more detail. Finally, this article addresses limitations and considerations regarding clinical machine learning applications in neurology.

Differentiation of benign and malignant vertebral fractures using a convolutional neural network to extract CT-based texture features

PURPOSE

To assess the diagnostic performance of three-dimensional (3D) CT-based texture features (TFs) using a convolutional neural network (CNN)-based framework to differentiate benign (osteoporotic) and malignant vertebral fractures (VFs).

METHODS

A total of 409 patients who underwent routine thoracolumbar spine CT at two institutions were included. VFs were categorized as benign or malignant using either biopsy or imaging follow-up of at least three months as standard of reference. Automated detection, labelling, and segmentation of the vertebrae were performed using a CNN-based framework ( https://anduin.bonescreen.de ). Eight TFs were extracted: Varianceglobal, Skewnessglobal, energy, entropy, short-run emphasis (SRE), long-run emphasis (LRE), run-length non-uniformity (RLN), and run percentage (RP). Multivariate regression models adjusted for age and sex were used to compare TFs between benign and malignant VFs.

RESULTS

Skewnessglobal showed a significant difference between the two groups when analyzing fractured vertebrae from T1 to L6 (benign fracture group: 0.70 [0.64-0.76]; malignant fracture group: 0.59 [0.56-0.63]; and p = 0.017), suggesting a higher skewness in benign VFs compared to malignant VFs.

CONCLUSION

Three-dimensional CT-based global TF skewness assessed using a CNN-based framework showed significant difference between benign and malignant thoracolumbar VFs and may therefore contribute to the clinical diagnostic work-up of patients with VFs.

Convolutional Neural Network for Fully Automated Cerebellar Volumetry in Children in Comparison to Manual Segmentation and Developmental Trajectory of Cerebellar Volumes

The purpose of this study was to develop a fully automated and reliable volumetry of the cerebellum of children during infancy and childhood using deep learning algorithms in comparison to manual segmentation. In addition, the clinical usefulness of measuring the cerebellar volume is shown. One hundred patients (0 to 16.3 years old) without infratentorial signal abnormalities on conventional MRI were retrospectively selected from our pool of pediatric MRI examinations. Based on a routinely acquired 3D T1-weighted magnetization prepared rapid gradient echo (MPRAGE) sequence, the cerebella were manually segmented using ITK-SNAP. The data set of all 100 cases was divided into four splits (four-fold cross-validation) to train the network (NN) to delineate the boundaries of the cerebellum. First, the accuracy of the newly created neural network was compared with the manual segmentation. Secondly, age-related volume changes were investigated. Our trained NN achieved an excellent Spearman correlation coefficient of 0.99, a Dice Coefficient of 95.0 ± 2.1%, and an intersection over union (IoU) of 90.6 ± 3.8%. Cerebellar volume increased continuously with age, showing an exponentially rapid growth within the first year of life. Using a convolutional neural network, it was possible to achieve reliable, fully automated cerebellar volume measurements in childhood and infancy, even when based on a relatively small cohort. In this preliminary study, age-dependent cerebellar volume changes could be acquired.

Lumbar Spinal Canal Segmentation in Cases with Lumbar Stenosis Using Deep-U-Net Ensembles

BACKGROUND

Narrowing of the lumbar spinal canal, or lumbar stenosis (LS), may cause debilitating radicular pain or muscle weakness. It is the most frequent indication for spinal surgery in the elderly population. Modern diagnosis relies on magnetic resonance imaging and its inherently subjective interpretation. Diagnostic rigor, accuracy, and speed may be improved by automation. In this work, we aimed to determine whether a deep-U-Net ensemble trained to segment spinal canals on a heterogeneous mix of clinical data is comparable to radiologists' segmentation of these canals in patients with LS.

METHODS

The deep U-nets were trained on spinal canals segmented by physicians on 100 axial T2 lumbar magnetic resonance imaging selected randomly from our institutional database. Test data included a total of 279 elderly patients with LS that were separate from the training set.

RESULTS

Machine-generated segmentations (MA) were qualitatively similar to expert-generated segmentations (ME1, ME2). Machine- and expert-generated segmentations were quantitatively similar, as evidenced by Dice scores (MA vs. ME1: 0.88 ± 0.04, MA vs. ME2: 0.89 ± 0.04), the Hausdorff distance (MA vs. ME1: 11.7 mm ± 13.8, MA vs. ME2: 13.1 mm ± 16.3), and average surface distance (MAvs. ME1: 0.18 mm ± 0.13, MA vs. ME2 0.18 mm ± 0.16) metrics. These metrics are comparable to inter-rater variation (ME1 vs. ME2 Dice scores: 0.94 ± 0.02, the Hausdorff distances: 9.3 mm ± 15.6, average surface distances: 0.08 mm ± 0.09).

CONCLUSION

We conclude that machine learning algorithms can segment lumbar spinal canals in LS patients, and automatic delineations are both qualitatively and quantitatively comparable to expert-generated segmentations.

Validating ASHS-T1 automated entorhinal and transentorhinal cortical segmentation in Alzheimer's disease

The current study aimed to validate entorhinal and transentorhinal cortical volumes measured by the automated segmentation tool Automatic Segmentation of Hippocampal Subfields (ASHS-T1). The study sample comprised 34 healthy controls (HCs), 37 individuals with amnestic mild cognitive impairment (aMCI), and 29 individuals with Alzheimer's disease (AD) dementia from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. Entorhinal and transentorhinal cortical volumes were assessed using ASHS-T1, manual segmentation, as well as a widely used automated segmentation tool, FreeSurfer v6.0.1. Mean differences, intraclass correlation coefficients, and Bland-Altman plots were computed. ASHS-T1 tended to underestimate entorhinal and transentorhinal cortical volumes relative to manual segmentation and FreeSurfer. There was variable consistency and low agreement between ASHS-T1 and manual segmentation volumes. There was low-to-moderate consistency and low agreement between ASHS-T1 and FreeSurfer volumes. There was a trend toward higher consistency and agreement for the entorhinal cortex in the aMCI and AD groups compared to the HC group. Despite the differences in volume measurements, ASHS-T1 was sensitive to entorhinal and transentorhinal cortical atrophy in both early and late disease stages. Based on the current study, ASHS-T1 appears to be a promising tool for automated entorhinal and transentorhinal cortical volume measurement in individuals with likely underlying AD.

Automated contouring and statistical process control for plan quality in a breast clinical trial

Background and purpose

Automatic review of breast plan quality for clinical trials is time-consuming and has some unique challenges due to the lack of target contours for some planning techniques. We propose using an auto-contouring model and statistical process control to independently assess planning consistency in retrospective data from a breast radiotherapy clinical trial.

Materials and methods

A deep learning auto-contouring model was created and tested quantitatively and qualitatively on 104 post-lumpectomy patients' computed tomography images (nnUNet; train/test: 80/20). The auto-contouring model was then applied to 127 patients enrolled in a clinical trial. Statistical process control was used to assess the consistency of the mean dose to auto-contours between plans and treatment modalities by setting control limits within three standard deviations of the data's mean. Two physicians reviewed plans outside the limits for possible planning inconsistencies.

Results

Mean Dice similarity coefficients comparing manual and auto-contours was above 0.7 for breast clinical target volume, supraclavicular and internal mammary nodes. Two radiation oncologists scored 95% of contours as clinically acceptable. The mean dose in the clinical trial plans was more variable for lymph node auto-contours than for breast, with a narrower distribution for volumetric modulated arc therapy than for 3D conformal treatment, requiring distinct control limits. Five plans (5%) were flagged and reviewed by physicians: one required editing, two had clinically acceptable variations in planning, and two had poor auto-contouring.

Conclusions

An automated contouring model in a statistical process control framework was appropriate for assessing planning consistency in a breast radiotherapy clinical trial.

A whole-body diffusion MRI normal atlas: development, evaluation and initial use

BACKGROUND

Statistical atlases can provide population-based descriptions of healthy volunteers and/or patients and can be used for region- and voxel-based analysis. This work aims to develop whole-body diffusion atlases of healthy volunteers scanned at 1.5T and 3T. Further aims include evaluating the atlases by establishing whole-body Apparent Diffusion Coefficient (ADC) values of healthy tissues and including healthy tissue deviations in an automated tumour segmentation task.

METHODS

Multi-station whole-body Diffusion Weighted Imaging (DWI) and water-fat Magnetic Resonance Imaging (MRI) of healthy volunteers (n = 45) were acquired at 1.5T (n = 38) and/or 3T (n = 29), with test-retest imaging for five subjects per scanner. Using deformable image registration, whole-body MRI data was registered and composed into normal atlases. Healthy tissue ADCmean was manually measured for ten tissues, with test-retest percentage Repeatability Coefficient (%RC), and effect of age, sex and scanner assessed. Voxel-wise whole-body analyses using the normal atlases were studied with ADC correlation analyses and an automated tumour segmentation task. For the latter, lymphoma patient MRI scans (n = 40) with and without information about healthy tissue deviations were entered into a 3D U-Net architecture.

RESULTS

Sex- and Body Mass Index (BMI)-stratified whole-body high b-value DWI and ADC normal atlases were created at 1.5T and 3T. %RC of healthy tissue ADCmean varied depending on tissue assessed (4-48% at 1.5T, 6-70% at 3T). Scanner differences in ADCmean were visualised in Bland-Altman analyses of dually scanned subjects. Sex differences were measurable for liver, muscle and bone at 1.5T, and muscle at 3T. Volume of Interest (VOI)-based multiple linear regression, and voxel-based correlations in normal atlas space, showed that age and ADC were negatively associated for liver and bone at 1.5T, and positively associated with brain tissue at 1.5T and 3T. Adding voxel-wise information about healthy tissue deviations in an automated tumour segmentation task gave numerical improvements in the segmentation metrics Dice score, sensitivity and precision.

CONCLUSIONS

Whole-body DWI and ADC normal atlases were created at 1.5T and 3T, and applied in whole-body voxel-wise analyses.

Comparison of the morphological characteristics of the choroidal sublayer between idiopathic macular holes and epiretinal membranes with automatic analysis

PURPOSE

To compare the choroidal sublayer morphologic features between idiopathic macular hole (IMH) and idiopathic epiretinal membrane (iERM) on spectral-domain optical coherent tomography (SD-OCT) using an automatic segmentation model.

METHODS

Thirty-three patients with idiopathic IMHs and 44 with iERMs who underwent vitrectomies were involved. The enhanced depth imaging mode of SD-OCT was used to obtain the B-scan image after single line scanning of the macular fovea. The choroidal sublayer automatic analysis model divides the choroidal into the choroidal large vessel layer, the middle vessel layer and the small vessel layer (LVCL, MVCL and SVCL, respectively) and calculates the choroidal thickness (overall, LVCL, MVCL and SVCL) and vascular index (overall, LVCL, MVCL and SVCL). The morphological characteristics of the choroidal sublayer in the ERM eyes and the IMH eyes were compared.

RESULTS

The mean choroidal thickness in the macular centre of the IMH eyes was significantly thinner than that of the ERM eyes (206.35 ± 81.72 vs. 273.33 ± 82.31 μm; P < 0.001). The analysis of the choroidal sublayer showed that the MVCL and SVCL macular centres and 0.5-1.5 mm of the nasal and temporal macula were significantly thinner in the IMH eyes than in the ERM eyes (P < 0.05), and there was a difference in the macular centre of the LVCL between the two groups (P < 0.05). In contrast, the choroidal vascular index of the macular centre in the IMH eyes was significantly higher than that in iERM eyes (0.2480 ± 0.0536 vs. 0.2120 ± 0.0616; P < 0.05). There was no significant difference in the CVI of other parts of the macula, the LVCL or MVCL between the two groups.

CONCLUSION

The choroidal thickness of the IMH eyes was significantly thinner than that of the iERM eyes, which was mainly observed in 3 mm of the macular centre and the MVCL and SVCL layers of the choroid. The choroidal vascular index of the IMH eyes was higher than that of the iERM eyes. These findings suggest that the choroid may be involved in the pathogenesis of IMH and iERM.

Automatic Detection of Visceral Arterial Aneurysms on Computed Tomography Angiography Using Artificial Intelligence Based Segmentation of the Vascular System

Introduction

Visceral arterial aneurysms (VAAs) are life threatening. Due to the paucity of symptoms and rarity of the disease, VAAs are underdiagnosed and underestimated. Artificial intelligence (AI) offers new insights into segmentation of the vascular system, and opportunities to better detect VAAs. This pilot study aimed to develop an AI based method to automatically detect VAAs from computed tomography angiography (CTA).

Methods

A hybrid method combining a feature based expert system with a supervised deep learning algorithm (convolutional neural network) was used to enable fully automatic segmentation of the abdominal vascular tree. Centrelines were built and reference diameters of each visceral artery were calculated. An abnormal dilatation (VAAs) was defined as a substantial increase in diameter at the pixel of interest compared with the mean diameter of the reference portion. The automatic software provided 3D rendered images with a flag on the identified VAA areas. The performance of the method was tested in a dataset of 33 CTA scans and compared with the ground truth provided by two human experts.

Results

Forty-three VAAs were identified by human experts (32 in the coeliac trunk branches, eight in the superior mesenteric artery, one in the left renal, and two in the right renal arteries). The automatic system accurately detected 40 of the 43 VAAs, with a sensitivity of 0.93 and a positive predictive value of 0.51. The mean number of flag areas per CTA was 3.5 ± 1.5 and they could be reviewed and checked by a human expert in less than 30 seconds per CTA.

Conclusion

Although the specificity needs to be improved, this study demonstrates the potential of an AI based automatic method to develop new tools to improve screening and detection of VAAs by automatically attracting clinicians' attention to suspicious dilatations of the visceral arteries.

BOUNTI: Brain vOlumetry and aUtomated parcellatioN for 3D feTal MRI

Fetal MRI is widely used for quantitative brain volumetry studies. However, currently, there is a lack of universally accepted protocols for fetal brain parcellation and segmentation. Published clinical studies tend to use different segmentation approaches that also reportedly require significant amounts of time-consuming manual refinement. In this work, we propose to address this challenge by developing a new robust deep learning-based fetal brain segmentation pipeline for 3D T2w motion corrected brain images. At first, we defined a new refined brain tissue parcellation protocol with 19 regions-of-interest using the new fetal brain MRI atlas from the Developing Human Connectome Project. This protocol design was based on evidence from histological brain atlases, clear visibility of the structures in individual subject 3D T2w images and the clinical relevance to quantitative studies. It was then used as a basis for developing an automated deep learning brain tissue parcellation pipeline trained on 360 fetal MRI datasets with different acquisition parameters using semi-supervised approach with manually refined labels propagated from the atlas. The pipeline demonstrated robust performance for different acquisition protocols and GA ranges. Analysis of tissue volumetry for 390 normal participants (21-38 weeks gestational age range), scanned with three different acquisition protocols, did not reveal significant differences for major structures in the growth charts. Only minor errors were present in < 15% of cases thus significantly reducing the need for manual refinement. In addition, quantitative comparison between 65 fetuses with ventriculomegaly and 60 normal control cases were in agreement with the findings reported in our earlier work based on manual segmentations. These preliminary results support the feasibility of the proposed atlas-based deep learning approach for large-scale volumetric analysis. The created fetal brain volumetry centiles and a docker with the proposed pipeline are publicly available online at https://hub.docker.com/r/fetalsvrtk/segmentation (tag brain_bounti_tissue).

Quantifying numerical and spatial reliability of hippocampal and amygdala subdivisions in FreeSurfer

On-going, large-scale neuroimaging initiatives can aid in uncovering neurobiological causes and correlates of poor mental health, disease pathology, and many other important conditions. As projects grow in scale with hundreds, even thousands, of individual participants and scans collected, quantification of brain structures by automated algorithms is becoming the only truly tractable approach. Here, we assessed the spatial and numerical reliability for newly deployed automated segmentation of hippocampal subfields and amygdala nuclei in FreeSurfer 7. In a sample of participants with repeated structural imaging scans (N = 928), we found numerical reliability (as assessed by intraclass correlations, ICCs) was reasonable. Approximately 95% of hippocampal subfields had "excellent" numerical reliability (ICCs ≥ 0.90), while only 67% of amygdala subnuclei met this same threshold. In terms of spatial reliability, 58% of hippocampal subfields and 44% of amygdala subnuclei had Dice coefficients ≥ 0.70. Notably, multiple regions had poor numerical and/or spatial reliability. We also examined correlations between spatial reliability and person-level factors (e.g., participant age; T1 image quality). Both sex and image scan quality were related to variations in spatial reliability metrics. Examined collectively, our work suggests caution should be exercised for a few hippocampal subfields and amygdala nuclei with more variable reliability.

Automated hemodynamic assessment for cranial 4D flow MRI

Cranial 4D flow MRI post-processing typically involves manual user interaction which is time-consuming and associated with poor repeatability. The primary goal of this study is to develop a robust quantitative velocity tool (QVT) that utilizes threshold-based segmentation techniques to improve segmentation quality over prior approaches based on centerline processing schemes (CPS) that utilize k-means clustering segmentation. This tool also includes an interactive 3D display designed for simplified vessel selection and automated hemodynamic visualization and quantification. The performances of QVT and CPS were compared in vitro in a flow phantom and in vivo in 10 healthy participants. Vessel segmentations were compared with ground-truth computed tomography in vitro (29 locations) and manual segmentation in vivo (13 locations) using linear regression. Additionally, QVT and CPS MRI flow rates were compared to perivascular ultrasound flow in vitro using linear regression. To assess internal consistency of flow measures in vivo, conservation of flow was assessed at vessel junctions using linear regression and consistency of flow along vessel segments was analyzed by fitting a Gaussian distribution to a histogram of normalized flow values. Post-processing times were compared between the QVT and CPS using paired t-tests. Vessel areas segmented in vitro (CPS: slope = 0.71, r = 0.95 and QVT: slope = 1.03, r = 0.95) and in vivo (CPS: slope = 0.61, r = 0.96 and QVT: slope = 0.93, r = 0.96) were strongly correlated with ground-truth area measurements. However, CPS (using k-means segmentation) consistently underestimated vessel areas. Strong correlations were observed between QVT and ultrasound flow (slope = 0.98, r = 0.96) as well as flow at junctions (slope = 1.05, r = 0.98). Mean and standard deviation of flow along vessel segments was 9.33e-16 ± 3.05%. Additionally, the QVT demonstrated excellent interobserver agreement and significantly reduced post-processing by nearly 10 min (p < 0.001). By completely automating post-processing and providing an easy-to-use 3D visualization interface for interactive vessel selection and hemodynamic quantification, the QVT offers an efficient, robust, and repeatable means to analyze cranial 4D flow MRI. This software is freely available at: https://github.com/uwmri/QVT.

High-throughput segmentation, data visualization, and analysis of sea star skeletal networks

The remarkably complex skeletal systems of the sea stars (Echinodermata, Asteroidea), consisting of hundreds to thousands of individual elements (ossicles), have intrigued investigators for more than 150 years. While the general features and structural diversity of isolated asteroid ossicles have been well documented in the literature, the task of mapping the spatial organization of these constituent skeletal elements in a whole-animal context represents an incredibly laborious process, and as such, has remained largely unexplored. To address this unmet need, particularly in the context of understanding structure-function relationships in these complex skeletal systems, we present an integrated approach that combines micro-computed tomography, semi-automated ossicle segmentation, data visualization tools, and the production of additively manufactured tangible models to reveal biologically relevant structural data that can be rapidly analyzed in an intuitive manner. In the present study, we demonstrate this high-throughput workflow by segmenting and analyzing entire skeletal systems of the giant knobby star, Pisaster giganteus, at four different stages of growth. The in-depth analysis, presented herein, provides a fundamental understanding of the three-dimensional skeletal architecture of the sea star body wall, the process of skeletal maturation during growth, and the relationship between skeletal organization and morphological characteristics of individual ossicles. The widespread implementation of this approach for investigating other species, subspecies, and growth series has the potential to fundamentally improve our understanding of asteroid skeletal architecture and biodiversity in relation to mobility, feeding habits, and environmental specialization in this fascinating group of echinoderms.

Automatic Segmentation of Parkinson Disease Therapeutic Targets Using Nonlinear Registration and Clinical MR Imaging: Comparison of Methodology, Presence of Disease, and Quality Control

INTRODUCTION

Accurate and precise delineation of the globus pallidus pars interna (GPi) and subthalamic nucleus (STN) is critical for the clinical treatment and research of Parkinson's disease (PD). Automated segmentation is a developing technology which addresses limitations of visualizing deep nuclei on MR imaging and standardizing their definition in research applications. We sought to compare manual segmentation with three workflows for template-to-patient nonlinear registration providing atlas-based automatic segmentation of deep nuclei.

METHODS

Bilateral GPi, STN, and red nucleus (RN) were segmented for 20 PD and 20 healthy control (HC) subjects using 3T MRIs acquired for clinical purposes. The automated workflows used were an option available in clinical practice and two common research protocols. Quality control (QC) was performed on registered templates via visual inspection of readily discernible brain structures. Manual segmentation using T1, proton density, and T2 sequences was used as "ground truth" data for comparison. Dice similarity coefficient (DSC) was used to assess agreement between segmented nuclei. Further analysis was done to compare the influences of disease state and QC classifications on DSC.

RESULTS

Automated segmentation workflows (CIT-S, CRV-AB, and DIST-S) had the highest DSC for the RN and lowest for the STN. Manual segmentations outperformed automated segmentation for all workflows and nuclei; however, for 3/9 workflows (CIT-S STN, CRV-AB STN, and CRV-AB GPi) the differences were not statically significant. HC and PD only showed significant differences in 1/9 comparisons (DIST-S GPi). QC classification only demonstrated significantly higher DSC in 2/9 comparisons (CRV-AB RN and GPi).

CONCLUSION

Manual segmentations generally performed better than automated segmentations. Disease state does not appear to have a significant effect on the quality of automated segmentations via nonlinear template-to-patient registration. Notably, visual inspection of template registration is a poor indicator of the accuracy of deep nuclei segmentation. As automatic segmentation methods continue to evolve, efficient and reliable QC methods will be necessary to support safe and effective integration into clinical workflows.

Deep Learning-Based Computed Tomography Image Standardization to Improve Generalizability of Deep Learning-Based Hepatic Segmentation

OBJECTIVE

We aimed to investigate whether image standardization using deep learning-based computed tomography (CT) image conversion would improve the performance of deep learning-based automated hepatic segmentation across various reconstruction methods.

MATERIALS AND METHODS

We collected contrast-enhanced dual-energy CT of the abdomen that was obtained using various reconstruction methods, including filtered back projection, iterative reconstruction, optimum contrast, and monoenergetic images with 40, 60, and 80 keV. A deep learning based image conversion algorithm was developed to standardize the CT images using 142 CT examinations (128 for training and 14 for tuning). A separate set of 43 CT examinations from 42 patients (mean age, 10.1 years) was used as the test data. A commercial software program (MEDIP PRO v2.0.0.0, MEDICALIP Co. Ltd.) based on 2D U-NET was used to create liver segmentation masks with liver volume. The original 80 keV images were used as the ground truth. We used the paired t-test to compare the segmentation performance in the Dice similarity coefficient (DSC) and difference ratio of the liver volume relative to the ground truth volume before and after image standardization. The concordance correlation coefficient (CCC) was used to assess the agreement between the segmented liver volume and ground-truth volume.

RESULTS

The original CT images showed variable and poor segmentation performances. The standardized images achieved significantly higher DSCs for liver segmentation than the original images (DSC [original, 5.40%-91.27%] vs. [standardized, 93.16%-96.74%], all P < 0.001). The difference ratio of liver volume also decreased significantly after image conversion (original, 9.84%-91.37% vs. standardized, 1.99%-4.41%). In all protocols, CCCs improved after image conversion (original, -0.006-0.964 vs. standardized, 0.990-0.998).

CONCLUSION

Deep learning-based CT image standardization can improve the performance of automated hepatic segmentation using CT images reconstructed using various methods. Deep learning-based CT image conversion may have the potential to improve the generalizability of the segmentation network.

Novel measures for the diagnosis of hepatic steatosis using contrast-enhanced computer tomography images

PURPOSE

Hepatic steatosis is often diagnosed non-invasively. Various measures and accompanying diagnostic thresholds based on contrast-enhanced CT and virtual non-contrast images have been proposed. We compare these established criteria to novel and fully automated measures.

METHOD

CT data sets of 197 patients were analyzed. Regions of interest (ROIs) were manually drawn for the liver, spleen, portal vein, and aorta to calculate four established measures of liver-fat. Two novel measures capturing the deviation between the empirical distributions of HU measurements across all voxels within the liver and spleen were calculated. These measures were calculated with both manual ROIs and using fully automated organ segmentations. Agreement between the different measures was evaluated using correlational analysis, as well as their ability to discriminate between fatty and healthy liver.

RESULTS

Established and novel measures of fatty liver were at a high level of agreement. Novel methods were statistically indistinguishable from the established ones when taking established diagnostic thresholds or physicians' diagnoses as ground truth and this high performance level persisted for automatically selected ROIs.

CONCLUSION

Automatically generated organ segmentations led to comparable results as manual ROIs, suggesting that the implementation of automated methods can prove to be a valuable tool for incidental diagnosis. Differences in the distribution of HU measurements across voxels between liver and spleen can serve as surrogate markers for the liver-fat-content. Novel measures do not exhibit a measurable disadvantage over established methods based on simpler measures such as across-voxel averages in a population with low incidence of fatty liver.

Automated assessment of the smoothness of retinal layers in optical coherence tomography images using a machine learning algorithm

Quantifying the smoothness of different layers of the retina can potentially be an important and practical biomarker in various pathologic conditions like diabetic retinopathy. The purpose of this study is to develop an automated machine learning algorithm which uses support vector regression method with wavelet kernel and automatically segments two hyperreflective retinal layers (inner plexiform layer (IPL) and outer plexiform layer (OPL)) in 50 optical coherence tomography (OCT) slabs and calculates the smoothness index (SI). The Bland-Altman plots, mean absolute error, root mean square error and signed error calculations revealed a modest discrepancy between the manual approach, used as the ground truth, and the corresponding automated segmentation of IPL/ OPL, as well as SI measurements in OCT slabs. It was concluded that the constructed algorithm may be employed as a reliable, rapid and convenient approach for segmenting IPL/OPL and calculating SI in the appropriate layers.

Open-source pipeline for automatic segmentation and microstructural analysis of murine knee subchondral bone

μCT images are commonly analysed to assess changes in bone density and microstructure in preclinical murine models. Several platforms provide automated analysis of bone microstructural parameters from volumetric regions of interest (ROI). However, segmentation of the regions of subchondral bone to create the volumetric ROIs remains a manual and time-consuming task. This study aimed to develop an automated end-to-end pipeline, combining segmentation and microstructural analysis, to evaluate subchondral bone in the mouse proximal knee.

METHODS

A segmented dataset of μCT scans from 62 knees (healthy and arthritic) from 10-week male C57BL/6 mice was used to train a U-Net type architecture to automate segmentation of the subchondral trabecular bone. These segmentations were used in tandem with the original scans as input for microstructural analysis along with thresholded trabecular bone. Manually and U-Net segmented ROIs were fed into two available pipelines for microstructural analysis: the ITKBoneMorphometry library and CTan (SKYSCAN). Outcome parameters were compared between pipelines, including: bone volume (BV), total volume (TV), BV/TV, trabecular number (TbN), trabecular thickness (TbTh), trabecular separation (TbSp), and bone surface density (BSBV).

RESULTS

There was good agreement for all bone measures comparing the manual and U-Net pipelines utilizing ITK (R = 0.88-0.98) and CTAn (R = 0.91-0.98). ITK and CTAn showed good agreement for BV, TV, BV/TV, TbTh and BSBV (R = 0.9-0.98). However, limited agreement was seen between TbN (R = 0.73) and TbSb (R = 0.59) due to methodological differences in how spacing is evaluated. Microstructural parameters generated from manual and automatic segmentations showed high correlation across all measures. Using the CTAn pipeline yielded strong R² values (0.83-0.96) and very strong agreement based on ICC (0.90-0.98). The ITK pipeline yielded similarly high R² values (0.91-0.96, except for TbN (0.77)), and ICC values (0.88-0.98). The automated segmentations yield lower average values for BV, TV and BV/TV (ranging from 14 % to 6.3 %), but differences were not found to be influenced by the mean ROI values.

CONCLUSIONS

This integrated pipeline seamlessly automated both segmentation and quantification of the proximal tibia subchondral bone microstructure. This automated pipeline allows the analysis of large volumes of data, and its open-source nature may enable the standardization of microstructural analysis of trabecular bone across different research groups.

Validation of skeletal muscle and adipose tissue measurements using a fully automated body composition analysis neural network versus a semi-automatic reference program with human correction in patients with lung cancer

Rationale and objectives

To validate skeletal muscle and adipose tissues cross sectional area (CSA) and densities between a fully automated neural network (test program) and a semi-automated program requiring human correction (reference program) for lumbar 1 (L1) and lumbar 2 (L2) CT scans in patients with lung cancer.

Materials and methods

Agreement between the reference and test programs was measured using Dice-similarity coefficient (DSC) and Bland-Altman plots with limits of agreement within 1.96 standard deviation.

Results

A total of 49 L1 and 47 L2 images were analyzed from patients with lung cancer (mean age = 70.51 ± 9.48 years; mean BMI = 27.45 ± 6.06 kg/m²; 71% female, 55% self-identified as Black and 96% as non-Hispanic ethnicity). The DSC indicates excellent overlap (>0.944) or agreement between the two measurement methods for muscle, visceral adipose tissue (VAT) and subcutaneous adipose tissue (SAT) CSA and all tissue densities at L1 and L2. The DSC was lowest for intermuscular adipose tissue (IMAT) CSA at L1 (0.889) and L2 (0.919).

Conclusion

The use of a fully automated neural network to analyze body composition at L1 and L2 in patients with lung cancer is valid for measuring skeletal muscle and adipose tissue CSA and densities when compared to a reference program. Further validation in a more diverse sample and in different disease and health states is warranted to increase the generalizability of the test program at L1 and L2.

A combined active shape and mean appearance model for the reconstruction of segmental bone loss

This study investigates the novel combination of an active shape and mean appearance model to estimate missing bone geometry and density distribution from sparse inputs simulating segmental bone loss of the femoral diaphysis. An active shape Gaussian Process Morphable model was trained on healthy right femurs of South African males to model shape. The density distribution was approximated based on the mean appearance of computed tomography images from the training set. Estimations of diaphyseal resections were obtained by probabilistic fitting of the active shape model to sparse inputs consisting of proximal and distal femoral data on computed tomography images. The resulting shape estimates of the diaphyseal resections were then used to map the mean appearance model to the patients' missing bone geometry, constructing density estimations. In this way, resected bone surfaces were estimated with an average error of 2.24 (0.5) mm. Density distributions were approximated within 87 (0.7) % of the intensity of the original target images before the simulated segmental bone loss. These results fall within the acceptable tolerances required for surgical planning and reconstruction of long bone defects.

Clinical evaluation of automated segmentation for body composition analysis on abdominal L3 CT slices in polytrauma patients

INTRODUCTION

Sarcopenia is a muscle disease that involves loss of muscle strength and physical function and is associated with adverse health effects. Even though sarcopenia has attracted increasing attention in the literature, many research findings have not yet been translated into clinical practice. In this article, we aim to validate a deep learning neural network for automated segmentation of L3 CT slices and aim to explore the potential for clinical utilization of such a tool for clinical practice.

MATERIALS AND METHODS

A deep learning neural network was trained on a multi-centre collection of 3413 abdominal cancer surgery subjects to automatically segment muscle, subcutaneous and visceral adipose tissue at the L3 lumbar vertebral level. 536 Polytrauma subjects were used as an independent test set to show generalizability. The Dice Similarity Coefficient was calculated to validate the geometric similarity. Quantitative agreement was quantified using Bland-Altman's Limits of Agreement interval and Lin's Concordance Correlation Coefficient. To determine the potential clinical usability, randomly selected segmentation images were presented to a panel of experienced clinicians to rate on a Likert scale.

RESULTS

Deep learning results gave excellent agreement versus a human expert operator for all of the body composition indices, with Concordance Correlation Coefficient for skeletal muscle index of 0.92, Skeletal muscle radiation attenuation 0.94, Visceral Adipose Tissue index 0.99 and Subcutaneous Adipose Tissue Index 0.99. Triple-blinded visual assessment of segmentation by clinicians correlated only to the Dice coefficient, but had no association to quantitative body composition metrics which were accurate irrespective of clinicians' visual rating.

CONCLUSION

A deep learning method for automatic segmentation of truncal muscle, visceral and subcutaneous adipose tissue on individual L3 CT slices has been independently validated against expert human-generated results for an enlarged polytrauma registry dataset. Time efficiency, consistency and high accuracy relative to human experts suggest that quantitative body composition analysis with deep learning should is a promising tool for clinical application in a hospital setting.

Automated Segmentation of Autofluorescence Lesions in Stargardt Disease

OBJECTIVE

To train a deep learning (DL) algorithm to perform fully automated semantic segmentation of multiple autofluorescence lesion types in Stargardt disease.

DESIGN

Cross-sectional study with retrospective imaging data.

SUBJECTS

The study included 193 images from 193 eyes of 97 patients with Stargardt disease.

METHODS

Fundus autofluorescence images obtained from patient visits between 2013 and 2020 were annotated with ground-truth labels. Model training and evaluation were performed using fivefold cross-validation.

MAIN OUTCOMES MEASURES

Dice similarity coefficients, intraclass correlation coefficients, and Bland-Altman analyses comparing algorithm-predicted and grader-labeled segmentations.

RESULTS

The overall Dice similarity coefficient across all lesion classes was 0.78 (95% confidence interval [CI], 0.69-0.86). Dice coefficients were 0.90 (95% CI, 0.85-0.94) for areas of definitely decreased autofluorescence (DDAF), 0.55 (95% CI, 0.35-0.76) for areas of questionably decreased autofluorescence (QDAF), and 0.88 (95% CI, 0.73-1.00) for areas of abnormal background autofluorescence (ABAF). Intraclass correlation coefficients comparing the ground-truth and automated methods were 0.997 (95% CI, 0.996-0.998) for DDAF, 0.863 (95% CI, 0.823-0.895) for QDAF, and 0.974 (95% CI, 0.966-0.980) for ABAF.

CONCLUSIONS

A DL algorithm performed accurate segmentation of autofluorescence lesions in Stargardt disease, demonstrating the feasibility of fully automated segmentation as an alternative to manual or semiautomated labeling methods.

Deep learning-based automated segmentation of resection cavities on postsurgical epilepsy MRI

Accurate segmentation of surgical resection sites is critical for clinical assessments and neuroimaging research applications, including resection extent determination, predictive modeling of surgery outcome, and masking image processing near resection sites. In this study, an automated resection cavity segmentation algorithm is developed for analyzing postoperative MRI of epilepsy patients and deployed in an easy-to-use graphical user interface (GUI) that estimates remnant brain volumes, including postsurgical hippocampal remnant tissue. This retrospective study included postoperative T1-weighted MRI from 62 temporal lobe epilepsy (TLE) patients who underwent resective surgery. The resection site was manually segmented and reviewed by a neuroradiologist (JMS). A majority vote ensemble algorithm was used to segment surgical resections, using 3 U-Net convolutional neural networks trained on axial, coronal, and sagittal slices, respectively. The algorithm was trained using 5-fold cross validation, with data partitioned into training (N = 27) testing (N = 9), and validation (N = 9) sets, and evaluated on a separate held-out test set (N = 17). Algorithm performance was assessed using Dice-Sørensen coefficient (DSC), Hausdorff distance, and volume estimates. Additionally, we deploy a fully-automated, GUI-based pipeline that compares resection segmentations with preoperative imaging and reports estimates of resected brain structures. The cross-validation and held-out test median DSCs were 0.84 ± 0.08 and 0.74 ± 0.22 (median ± interquartile range) respectively, which approach inter-rater reliability between radiologists (0.84-0.86) as reported in the literature. Median 95 % Hausdorff distances were 3.6 mm and 4.0 mm respectively, indicating high segmentation boundary confidence. Automated and manual resection volume estimates were highly correlated for both cross-validation (r = 0.94, p < 0.0001) and held-out test subjects (r = 0.87, p < 0.0001). Automated and manual segmentations overlapped in all 62 subjects, indicating a low false negative rate. In control subjects (N = 40), the classifier segmented no voxels (N = 33), <50 voxels (N = 5), or a small volumes<0.5 cm3 (N = 2), indicating a low false positive rate that can be controlled via thresholding. There was strong agreement between postoperative hippocampal remnant volumes determined using automated and manual resection segmentations (r = 0.90, p < 0.0001, mean absolute error = 6.3 %), indicating that automated resection segmentations can permit quantification of postoperative brain volumes after epilepsy surgery. Applications include quantification of postoperative remnant brain volumes, correction of deformable registration, and localization of removed brain regions for network modeling.

Automatic quantification of white matter hyperintensities on T2-weighted fluid attenuated inversion recovery magnetic resonance imaging

White matter hyperintensities (WMH) are areas of increased signal visualized on T2-weighted fluid attenuated inversion recovery (FLAIR) brain magnetic resonance imaging (MRI) sequences. They are typically attributed to small vessel cerebrovascular disease in the context of aging. Among older adults, WMH are associated with risk of cognitive decline and dementia, stroke, and various other health outcomes. There has been increasing interest in incorporating quantitative WMH measurement as outcomes in clinical trials, observational research, and clinical settings. Here, we present a novel, fully automated, unsupervised detection algorithm for WMH segmentation and quantification. The algorithm uses a robust preprocessing pipeline, including brain extraction and a sample-specific mask that incorporates spatial information for automatic false positive reduction, and a half Gaussian mixture model (HGMM). The method was evaluated in 24 participants with varying degrees of WMH (4.9-78.6 cm³) from a community-based study of aging and dementia with dice coefficient, sensitivity, specificity, correlation, and bias relative to the ground truth manual segmentation approach performed by two expert raters. Results were compared with those derived from commonly used available WMH segmentation packages, including SPM lesion probability algorithm (LPA), SPM lesion growing algorithm (LGA), and Brain Intensity AbNormality Classification Algorithm (BIANCA). The HGMM algorithm derived WMH values that had a dice score of 0.87, sensitivity of 0.89, and specificity of 0.99 compared to ground truth. White matter hyperintensity volumes derived with HGMM were strongly correlated with ground truth values (r = 0.97, p = 3.9e-16), with no observable bias (-1.1 [-2.6, 0.44], p-value = 0.16). Our novel algorithm uniquely uses a robust preprocessing pipeline and a half-Gaussian mixture model to segment WMH with high agreement with ground truth for large scale studies of brain aging.

Automated tibiofemoral joint segmentation based on deeply supervised 2D-3D ensemble U-Net: Data from the Osteoarthritis Initiative

Improving longevity is one of the greatest achievements in humanity. Because of this, the population is growing older, and the ubiquity of knee osteoarthritis (OA) is on the rise. Nonetheless, the understanding and ability to investigate potential precursors of knee OA have been impeded by time-consuming and laborious manual delineation processes which are prone to poor reproducibility. A method for automatic segmentation of the tibiofemoral joint using magnetic resonance imaging (MRI) is presented in this work. The proposed method utilizes a deeply supervised 2D-3D ensemble U-Net, which consists of foreground class oversampling, deep supervision loss branches, and Gaussian weighted softmax score aggregation. It was designed, optimized, and tested on 507 3D double echo steady-state (DESS) MR volumes using a two-fold cross-validation approach. A state-of-the-art segmentation accuracy measured as Dice similarity coefficient (DSC) for the femur bone (98.6 ± 0.27%), tibia bone (98.8 ± 0.31%), femoral cartilage (90.3 ± 2.89%), and tibial cartilage (86.7 ± 4.07%) is achieved. Notably, the proposed method yields sub-voxel accuracy for an average symmetric surface distance (ASD) less than 0.36 mm. The model performance is not affected by the severity of radiographic osteoarthritis (rOA) grades or the presence of pathophysiological changes. The proposed method offers an accurate segmentation with high time efficiency (~62 s) per 3D volume, which is well suited for efficient processing and analysis of the large prospective cohorts of the Osteoarthritis Initiative (OAI).

Local MRI-based measures of thigh adipose tissue derived from fully automated deep convolutional neural network-based segmentation show a comparable responsiveness to bidirectional change in body weight as from quality controlled manual segmentation

BACKGROUND

Thigh intermuscular (IMF) and subcutaneous (SCF) fat are associated with joint function, inflammation and knee osteoarthritis. Fully automated segmentation from MRI is important to study the above relationship in larger cohorts. However, such algorithms are not clinically evaluated for longitudinal studies. Our aim was to evaluate a fully automated U-Net segmentation approach and its ability to detect longitudinal changes in thigh IMF and SCF during weight changes compared to manual segmentation.

METHODS

103 Osteoarthritis Initiative subjects, were studied, 52 with> 10% weight loss, and 51 with> 10% weight gain over 2-years. Longitudinal change in IMF and SCF were determined from baseline and year-2 axial thigh MRIs using U-Net segmentation. The standardised response mean (SRM) was used as measure of sensitivity to change.

RESULTS

The U-Net took substantially less time (single-slice MRI:< 1 s) and IMF and SCF showed very similar sensitivity to change as manual segmentation: With an average weight gain of + 14%, we observed an + 12% /+ 26% increase in IMF / SCF (SRM=0.99 /1.03) using the U-Net, compared with + 21% /+ 27% (SRM=0.60 /1.07) for manual segmentation. During an average weight loss of - 18%, we observed an - 14% /- 22% reduction in IMF /SCF (SRM = - 1.04 /-1.20) using the U-Net, compared with - 16% /- 22% (SRM = - 0.70 /-1.23) for manual segmentation.

CONCLUSION

U-Net segmentation replicates longitudinal changes of IMF and SCF associated with weight changes with a similar sensitivity to change as manual segmentation. This method is applicable to large databases for studying relationships between IMF and SCF and various disease conditions.

A comparison of automated atrophy measures across the frontotemporal dementia spectrum: Implications for trials

Background

Frontotemporal dementia (FTD) is a common cause of young onset dementia, and whilst there are currently no treatments, there are several promising candidates in development and early phase trials. Comprehensive investigations of neuroimaging markers of disease progression across the full spectrum of FTD disorders are lacking and urgently needed to facilitate these trials.

Objective

To investigate the comparative performance of multiple automated segmentation and registration pipelines used to quantify longitudinal whole-brain atrophy across the clinical, genetic and pathological subgroups of FTD, in order to inform upcoming trials about suitable neuroimaging-based endpoints.

Methods

Seventeen fully automated techniques for extracting whole-brain atrophy measures were applied and directly compared in a cohort of 226 participants who had undergone longitudinal structural 3D T1-weighted imaging. Clinical diagnoses were behavioural variant FTD (n = 56) and primary progressive aphasia (PPA, n = 104), comprising semantic variant PPA (n = 38), non-fluent variant PPA (n = 42), logopenic variant PPA (n = 18), and PPA-not otherwise specified (n = 6). 49 of these patients had either a known pathogenic mutation or postmortem confirmation of their underlying pathology. 66 healthy controls were included for comparison. Sample size estimates to detect a 30% reduction in atrophy (80% power; 0.05 significance) were computed to explore the relative feasibility of these brain measures as surrogate markers of disease progression and their ability to detect putative disease-modifying treatment effects.

Results

Multiple automated techniques showed great promise, detecting significantly increased rates of whole-brain atrophy (p<0.001) and requiring sample sizes of substantially less than 100 patients per treatment arm. Across the different FTD subgroups, direct measures of volume change consistently outperformed their indirect counterparts, irrespective of the initial segmentation quality. Significant differences in performance were found between both techniques and patient subgroups, highlighting the importance of informed biomarker choice based on the patient population of interest.

Conclusion

This work expands current knowledge and builds on the limited longitudinal investigations currently available in FTD, as well as providing valuable information about the potential of fully automated neuroimaging biomarkers for sporadic and genetic FTD trials.

Automatic segmentation of corneal deposits from corneal stromal dystrophy images via deep learning

BACKGROUND

Granular dystrophy is the most common stromal dystrophy. To perform automated segmentation of corneal stromal deposits, we trained and tested a deep learning (DL) algorithm from patients with corneal stromal dystrophy and compared its performance with human segmentation.

METHODS

In this retrospective cross-sectional study, we included slit-lamp photographs by sclerotic scatter from patients with corneal stromal dystrophy and real-world slit-lamp photographs via various techniques (diffuse illumination, tangential illumination, and sclerotic scatter). Our data set included 1007 slit-lamp photographs of semi-automatically generated handcraft masks on granular and linear lesions from corneal stromal dystrophy patients (806 for the training set and 201 for test set). For external test (140 photographs), we applied the DL algorithm and compared between automated and human segmentation. For performance, we estimated the intersection of union (IoU), global accuracy, and boundary F1 (BF) score for segmentation.

RESULTS

In 201 internal test set, IoU, global accuracy, and BF score with 95 % confidence Interval were 0.81 (0.79-0.82), 0.99 (0.98-0.99), and 0.93 (0.92-0.95), respectively. In 140 heterogenous external test set as a real-world data, those were 0.64 (0.61-0.67), 0.95 (0.94-0.96), and 0.70 (0.64-0.76) via DL algorithm and 0.56 (0.51-0.61), 0.95 (0.94-0.96), and 0.70 (0.65-0.74) via human rater, respectively.

CONCLUSIONS

We developed an automated segmentation DL algorithm for corneal stromal deposits in patients with corneal stromal dystrophy. Segmentation on corneal deposits was accurate via the DL algorithm in the well-controlled dataset and showed reasonable performance in a real-world setting. We suggest this automatic segmentation of corneal deposits helps to monitor the disease and can evaluate possible new treatments.

Machine Learning-Enabled Determination of Diffuseness of Brain Arteriovenous Malformations from Magnetic Resonance Angiography

The diffuseness of brain arteriovenous malformations (bAVMs) is a significant factor in surgical outcome evaluation and hemorrhagic risk prediction. However, there are still predicaments in identifying diffuseness, such as the judging variety resulting from different experience and difficulties in quantification. The purpose of this study was to develop a machine learning (ML) model to automatically identify the diffuseness of bAVM niduses using three-dimensional (3D) time-of-flight magnetic resonance angiography (TOF-MRA) images. A total of 635 patients with bAVMs who underwent TOF-MRA imaging were enrolled. Three experienced neuroradiologists delineated the bAVM lesions and identified the diffuseness on TOF-MRA images, which were considered the ground-truth reference. The U-Net-based segmentation model was trained to segment lesion areas. Eight mainstream ML models were trained through the radiomic features of segmented lesions to identify diffuseness, based on which an integrated model was built and yielded the best performance. In the test set, the Dice score, F2 score, precision, and recall for the segmentation model were 0.80 [0.72-0.84], 0.80 [0.71-0.86], 0.84 [0.77-0.93], and 0.82 [0.69-0.89], respectively. For the diffuseness identification model, the ensemble-based model was applied with an area under the Receiver-operating characteristic curves (AUC) of 0.93 (95% CI 0.87-0.99) in the training set. The AUC, accuracy, precision, recall, and F1 score for the diffuseness identification model were 0.95, 0.90, 0.81, 0.84, and 0.83, respectively, in the test set. The ML models showed good performance in automatically detecting bAVM lesions and identifying diffuseness. The method may help to judge the diffuseness of bAVMs objectively, quantificationally, and efficiently.

Improved real-time segmentation of Intravascular Ultrasound images using coordinate-aware fully convolutional networks

Segmentation of Intravascular Ultrasound (IVUS) images into Lumen and Media (interior and exterior) artery vessel walls is highly clinically relevant in the diagnosis and treatment of cardiovascular diseases such as atherosclerosis. When fused with position data, such segmentations also play a key role in reconstructing 3D representations of arteries. Automated segmentation in real-time is known to be a difficult image analysis problem, primarily due to artefacts commonly present in IVUS ultrasound images such as shadows, guide-wire effects, and side-branches. An additional challenge is the limited amount of expert labelled IVUS data, which limits the application of many well-performing deep learning models from other domains. To exploit the circular layered structure of the artery in B-Mode images, we propose a multi-class fully convolutional semantic segmentation network based on a minimal U-Net architecture augmented with learned translation dependence in the polar domain. The coordinate awareness in the multi-class segmentation allows the model to exploit relative spatial context about the interior and exterior vessel walls which are simply separable in polar coordinates. After training on 109 expert-labelled examples, our model significantly outperforms the state-of-the art in terms of mean Jaccard Measure (0.91 vs. 0.89) and Hausdorff distance (0.32 mm vs. 0.48 mm) on Media segmentation, and reaches equivalent performance on Lumen segmentation when evaluated on a standard publicly available dataset of 326 IVUS B-Mode images captured by 20 Mhz ultrasound probes. Using an order of magnitude fewer trainable parameters than the previous state-of-the-art, our model runs over 50 times faster and is able to execute in only 3 ms on a common GPU, achieving both leading accuracy and practical real-time performance.

Accuracy and longitudinal reproducibility of quantitative femorotibial cartilage measures derived from automated U-Net-based segmentation of two different MRI contrasts: data from the osteoarthritis initiative healthy reference cohort

OBJECTIVE

To evaluate the agreement, accuracy, and longitudinal reproducibility of quantitative cartilage morphometry from 2D U-Net-based automated segmentations for 3T coronal fast low angle shot (corFLASH) and sagittal double echo at steady-state (sagDESS) MRI.

METHODS

2D U-Nets were trained using manual, quality-controlled femorotibial cartilage segmentations available for 92 Osteoarthritis Initiative healthy reference cohort participants from both corFLASH and sagDESS (n = 50/21/21 training/validation/test-set). Cartilage morphometry was computed from automated and manual segmentations for knees from the test-set. Agreement and accuracy were evaluated from baseline visits (dice similarity coefficient: DSC, correlation analysis, systematic offset). The longitudinal reproducibility was assessed from year-1 and -2 follow-up visits (root-mean-squared coefficient of variation, RMSCV%).

RESULTS

Automated segmentations showed high agreement (DSC 0.89-0.92) and high correlations (r ≥ 0.92) with manual ground truth for both corFLASH and sagDESS and only small systematic offsets (≤ 10.1%). The automated measurements showed a similar test-retest reproducibility over 1 year (RMSCV% 1.0-4.5%) as manual measurements (RMSCV% 0.5-2.5%).

DISCUSSION

The 2D U-Net-based automated segmentation method yielded high agreement compared with manual segmentation and also demonstrated high accuracy and longitudinal test-retest reproducibility for morphometric analysis of articular cartilage derived from it, using both corFLASH and sagDESS.

Automated segmentation and description of the internal morphology of human permanent teeth by means of micro-CT

High-resolution micro-computed tomography is a powerful tool to analyze and visualize the internal morphology of human permanent teeth. It is increasingly used for investigation of epidemiological questions to provide the dentist with the necessary information required for successful endodontic treatment. The aim of the present paper was to propose an image processing method to automate parts of the work needed to fully describe the internal morphology of human permanent teeth. One hundred and four human teeth were scanned on a high-resolution micro-CT scanner using an automatic specimen changer. Python code in a Jupyter notebook was used to verify and process the scans, prepare the datasets for description of the internal morphology and to measure the apical region of the tooth. The presented method offers an easy, non-destructive, rapid and efficient approach to scan, check and preview tomographic datasets of a large number of teeth. It is a helpful tool for the detailed description and characterization of the internal morphology of human permanent teeth using automated segmentation by means of micro-CT with full reproducibility and high standardization.

Validation of automatic MRI hippocampal subfield segmentation by histopathological evaluation in patients with temporal lobe epilepsy

OBJECTIVE

The present study validates the results of automated hippocampal subfield segmentation with histopathology in epilepsy patients undergoing epilepsy surgery.

METHODS

We performed an automated hippocampal subfield segmentation on presurgical three-dimensional, T1-weighted magnetization Prepared Rapid Acquisition of Gradient Echoes Magnetic Resonance Imaging (MRI) data of 25 patients with unilateral mesial temporal lobe epilepsy due to hippocampal sclerosis (HS), using Freesurfer Version 6.0. The resulting volumes of cornu ammonis (CA) subfields CA1, CA2/3, CA4 and the dentate gyrus (DG) were compared to the histopathological cell count.

RESULTS

We found a significant correlation between histopathology in subregion CA2 and automated segmentation of subregion CA1 (p = 0.0062), CA2/3 (p = 0.004), CA4 (p = 0.0062) and the DG (p = 0.0054), between histopathology in CA3 and automated segmentation of CA1 (p = 0.0132), CA2/3 (p = 0.0004), CA4 (p = 0.0032) and the DG (p = 0.0037), as well as between histopathology in the DG and automated segmentation of CA1 (p = 0.0115), CA2/3 (p < 0.0001), CA4 (p < 0.0001) and the DG (p = 0.0001). The histopathological finding of HS type 1 could correctly be classified in all cases on MRI.

SIGNIFICANCE

The present study shows significant correlations between histopathological evaluation and results of the automated segmentation of the hippocampus, thereby validating the automated segmentation method. As the differential involvement of different hippocampal subfields may be associated with clinical parameters and the outcome after epilepsy surgery, the automated segmentation is also promising for prognostic purposes.

Focal cervical spinal stenosis causes mechanical strain on the entire cervical spinal cord tissue - A prospective controlled, matched-pair analysis based on phase-contrast MRI

BACKGROUND

Focally increased spinal cord motion at the level of cervical spinal stenosis has been revealed by phase-contrast MRI (PC-MRI).

OBJECTIVE

To investigate spinal cord motion among patients suffering of degenerative cervical myelopathy (DCM) across the entire cervical spine applying automated segmentation and standardized PC-MRI post-processing protocols.

METHODS

Prospective, matched-pair controlled trial on 29 patients with stenosis at C5/C6. MRI-protocol covering all cervical segments: 3D T2-SPACE, prospectively ECG-triggered sagittal PC-MRI. Segmentation by trained 3D hierarchical deep convolutional neural network and data processing were conducted via in-house software pipeline. Parameters per segment: maximum velocity, peak-to-peak (PTP)-amplitude, total displacement, PTP-amplitude^HB (PTP-amplitude per duration of heartbeat), and, for characterization of intraindividual alterations, the PTP-amplitude index between the cervical segments C3/C4-C7/T1 and C2/C3.

RESULTS

Spinal cord motion was increased at C4/C5, C5/C6 and C6/C7 among patients (all parameters, p < 0.001-0.025). The PTP-amplitude index revealed an increase from C3/C4 to C4/C5 (p = 0.002), C4/C5 to C5/C6 (p = 0.037) and a decrease from C5/C6 to C6/C7 and C6/C7 to C7/T1 (p < 0.001, each). This implied an up-building stretch on spinal cord tissue cranial and a mechanical compression caudal of the stenotic level. Furthermore, significant far range effects across the entire cervical spinal cord were observed (e.g. PTP-amplitude C2/C3 vs. C6/C7, p = 0.026) in contrast to controls (p = 1.00).

CONCLUSION

This study revealed the nature and extends of mechanical stress on the entire cervical spinal cord tissue due to focal stenosis. These pathophysiological alterations of spinal cord motion can be expected to be clinically relevant.

3D PBV-Net: An automated prostate MRI data segmentation method

Prostate cancer is one of the most common deadly diseases in men worldwide, which is seriously affecting people's life and health. Reliable and automated segmentation of the prostate gland in MRI data is exceptionally critical for diagnosis and treatment planning of prostate cancer. Although many automated segmentation methods have emerged, including deep learning based approaches, segmentation performance is still poor due to the large variability of image appearance, anisotropic spatial resolution, and imaging interference. This study proposes an automated prostate MRI data segmentation approach using bicubic interpolation with improved 3D V-Net (dubbed 3D PBV-Net). Considering the low-frequency components in the prostate gland, the bicubic interpolation is applied to preprocess the MRI data. On this basis, a 3D PBV-Net is developed to perform prostate MRI data segmentation. To illustrate the effectiveness of our approach, we evaluate the proposed 3D PBV-Net on two clinical prostate MRI data datasets, i.e., PROMISE 12 and TPHOH, with the manual delineations available as the ground truth. Our approach generates promising segmentation results, which have achieved 97.65% and 98.29% of average accuracy, 0.9613 and 0.9765 of Dice metric, 3.120 mm and 0.9382 mm of Hausdorff distance, and average boundary distance of 1.708, 0.7950 on PROMISE 12 and TPHOH datasets, respectively. Our method has effectively improved the accuracy of automated segmentation of the prostate MRI data and is promising to meet the accuracy requirements for telehealth applications.

Automatic multispectral MRI segmentation of human hippocampal subfields: an evaluation of multicentric test-retest reproducibility

Accurate and reproducible automated segmentation of human hippocampal subfields is of interest to study their roles in cognitive functions and disease processes. Multispectral structural MRI methods have been proposed to improve automated hippocampal subfield segmentation accuracy, but the reproducibility in a multicentric setting is, to date, not well characterized. Here, we assessed test-retest reproducibility of FreeSurfer 6.0 hippocampal subfield segmentations using multispectral MRI analysis pipelines (22 healthy subjects scanned twice, a week apart, at four 3T MRI sites). The harmonized MRI protocol included two 3D-T1, a 3D-FLAIR, and a high-resolution 2D-T2. After within-session T1 averaging, subfield volumes were segmented using three pipelines with different multispectral data: two longitudinal ("long_T1s" and "long_T1s_FLAIR") and one cross-sectional ("long_T1s_FLAIR_crossT2"). Volume reproducibility was quantified in magnitude (reproducibility error-RE) and space (DICE coefficient). RE was lower in all hippocampal subfields, except for hippocampal fissure, using the longitudinal pipelines compared to long_T1s_FLAIR_crossT2 (average RE reduction of 0.4-3.6%). Similarly, the longitudinal pipelines showed a higher spatial reproducibility (1.1-7.8% of DICE improvement) in all hippocampal structures compared to long_T1s_FLAIR_crossT2. Moreover, long_T1s_FLAIR provided a small but significant RE improvement in comparison to long_T1s (p = 0.015), whereas no significant DICE differences were found. In addition, structures with volumes larger than 200 mm³ had better RE (1-2%) and DICE (0.7-0.95) than smaller structures. In summary, our study suggests that the most reproducible hippocampal subfield FreeSurfer segmentations are derived from a longitudinal pipeline using 3D-T1s and 3D-FLAIR. Adapting a longitudinal pipeline to include high-resolution 2D-T2 may lead to further improvements.

Computer-aided quantification of non-contrast 3D black blood MRI as an efficient alternative to reference standard manual CT angiography measurements of abdominal aortic aneurysms

BACKGROUND

Non-contrast 3D black blood MRI is a promising tool for abdominal aortic aneurysm (AAA) surveillance, permitting accurate aneurysm diameter measurements needed for patient management.

PURPOSE

To evaluate whether automated AAA volume and diameter measurements obtained from computer-aided segmentation of non-contrast 3D black blood MRI are accurate, and whether they can supplant reference standard manual measurements from contrast-enhanced CT angiography (CTA).

MATERIALS AND METHODS

Thirty AAA patients (mean age, 71.9 ± 7.9 years) were recruited between 2014 and 2017. Participants underwent both non-contrast black blood MRI and CTA within 3 months of each other. Semi-automatic (computer-aided) MRI and CTA segmentations utilizing deformable registration methods were compared against manual segmentations of the same modality using the Dice similarity coefficient (DSC). AAA lumen and total aneurysm volumes and AAA maximum diameter, quantified automatically from these segmentations, were compared against manual measurements using Pearson correlation and Bland-Altman analyses. Finally, automated measurements from non-contrast 3D black blood MRI were evaluated against manual CTA measurements using the Wilcoxon test, Pearson correlation and Bland-Altman analyses.

RESULTS

Semi-automatic segmentations had excellent agreement with manual segmentations (lumen DSC: 0.91 ± 0.03 and 0.94 ± 0.03; total aneurysm DSC: 0.92 ± 0.02 and 0.94 ± 0.03, for black blood MRI and CTA, respectively). Automated volume and maximum diameter measurements also had excellent correlation to their manual counterparts for both black blood MRI (volume: r = 0.99, P < 0.001; diameter: r = 0.97, P < 0.001) and CTA (volume: r = 0.99, P < 0.001; diameter: r = 0.97, P < 0.001). Compared to manual CTA measurements, bias and limits of agreement (LOA) for automated MRI measurements (lumen volume: 1.49, [-4.19 7.17] cm³; outer wall volume: -2.46, [-14.05 9.13] cm³; maximal diameter: 0.08, [-6.51 6.67] mm) were largely equivalent to those of manual MRI measurements, particularly for maximum AAA diameter (lumen volume: 0.73, [-6.47 7.93] cm³; outer wall volume: 0.98, [-10.54 12.5] cm³; maximal diameter: 0.08, [-3.67 3.83] mm).

CONCLUSION

Semi-automatic segmentation of non-contrast 3D black blood MRI efficiently provides reproducible morphologic AAA assessment yielding accurate AAA diameters and volumes with no clinically relevant differences compared to either automatic or manual measurements based on CTA.

Automated Brain Region Segmentation for Single Cell Resolution Histological Images Based on Markov Random Field

The brain consists of massive regions with different functions and the precise delineation of brain region boundaries is important for brain region identification and atlas illustration. In this paper we propose a hierarchical Markov random field (MRF) model for brain region segmentation, where a MRF is applied to the downsampled low-resolution images and the result is used to initialize another MRF for the original high-resolution images. A fractional differential feature and a gray level co-occurrence matrix are extracted as the observed vector for the MRF and a new potential energy function, which can capture the spatial characteristic of brain regions, is proposed as well. A fuzzy entropy criterion is used to fine-tune the boundary from the hierarchical MRF model. We test the model both on synthetic images and real histological mouse brain images. The result suggests that the model can accurately identify target regions and even the whole mouse brain outline as a special case. An interesting observation is that the model cannot only segment regions with different cell density but also can segment regions with similar cell density and different cell morphology texture. Thus this model shows great potential for building the high-resolution 3D brain atlas.

A Valid and Precise Semiautomated Method for Quantifying Intermuscular Fat Intramuscular Fat in Lower Leg Magnetic Resonance Images

The accumulation of INTERmuscular fat and INTRAmuscular fat (IMF) has been a hallmark of individuals with diabetes, those with mobility impairments such as spinal cord injuries and is known to increase with aging. An elevated amount of IMF has been associated with fractures and frailty, but the imprecision of IMF measurement has so far limited the ability to observe more consistent clinical associations. Magnetic resonance imaging has been recognized as the gold standard for portraying these features, yet reliable methods for quantifying IMF on magnetic resonance imaging is far from standardized. Previous investigators used manual segmentation guided by histogram-based region-growing, but these techniques are subjective and have not demonstrated reliability. Others applied fuzzy classification, machine learning, and atlas-based segmentation methods, but each is limited by the complexity of implementation or by the need for a learning set, which must be established each time a new disease cohort is examined. In this paper, a simple convergent iterative threshold-optimizing algorithm was explored. The goal of the algorithm is to enable IMF quantification from plain fast spin echo (FSE) T1-weighted MR images or from water-saturated images. The algorithm can be programmed into Matlab easily, and is semiautomated, thus minimizing the subjectivity of threshold-selection. In 110 participants from 3 cohort studies, IMF area measurement demonstrated a high degree of reproducibility with errors well within the 5% benchmark for intraobserver, interobserver, and test-retest analyses; in contrast to manual segmentation which already yielded over 20% error for intraobserver analysis. This algorithm showed validity against manual segmentations (r > 0.85). The simplicity of this technique lends itself to be applied to fast spin echo images commonly ordered as part of standard of care and does not require more advanced fat-water separated images.

Cerebral artery segmentation based on magnetization-prepared two rapid acquisition gradient echo multi-contrast images in 7 Tesla magnetic resonance imaging

Cerebral artery segmentation plays an important role in the direct visualization of the human brain to obtain vascular system information. On ultra-high field magnetic resonance imaging, cerebral arteries appearing hyperintense on T1 weighted (T1w) images could be segmented from brain tissues such as gray and white matter. In this study, we propose an automated method to segment the cerebral arteries using multi-contrast images including T1w images of a magnetization-prepared two rapid acquisition gradient echo (MP2RAGE) sequence at 7 T. The proposed method, termed MP2rase-CA (MP2rage based RApid SEgmentation Cerebral Artery), employed a seed-based region-growing strategy and Frangi filtering as well as our brain tissue segmentation (MP2rase Brain Tissue). Time-of-flight (TOF) magnetic resonance angiography (MRA) images were obtained as a reference to evaluate the MP2rase-CA. We successfully performed vessel segmentations, from T1w MP2RAGE images, which mostly overlapped with the segmentations of large cerebral arteries from the TOF-MRA. We also investigated the effect of the large cerebral arteries on spatial transformation of anatomical images to standard coordinate space using vessel segmentation by MP2rase-CA. As a result, the T1w image without the cerebral arteries by MP2rase-CA showed better agreement with the standard atlas compared with the T1w image containing the arteries. In addition, voxel-based morphology showed significant differences between T1w images with and without cerebral arteries in brain areas nearby large arteries. Thus, because MP2rase-CA using MP2RAGE images can obtain brain tissue anatomical information as well as relatively large cerebral artery information without need for additional structure acquisition, it is useful and time saving for functional and structural studies.

Automated MicroCT-based bone and articular cartilage analysis using iterative shape averaging and atlas-based registration

Micro-computed tomography (μCT) and contrast-enhanced μCT are important tools for preclinical analysis of bone and articular cartilage (AC). Quantitative data from these modalities is highly dependent on the accuracy of tissue segmentations, which are often obtained via time-consuming manual contouring and are prone to inter- and intra-observer variability. Automated segmentation strategies could mitigate these issues, but few such approaches have been described in the context of μCT. Here, we validated a fully-automated strategy for bone and AC segmentation based on registration of an average tissue atlas. Femora from healthy and arthritic rats underwent μCT scanning, and epiphyseal trabecular bone and AC volumes were manually contoured by an expert. Average tissue atlases composed of 1, 3, 5, 10 and 20 pre-contoured training images (n = 10 atlases/group) were generated using iterative shape averaging and registered onto unknown images via affine and non-rigid registration. Atlas-based and expert-defined volumes for bone and AC were compared in terms of shape-based similarity metrics, as well as morphometric and densitometric parameters. Our results demonstrate that atlas-based registrations were capable of highly accurate and consistent segmentation. Atlases built from as few as 3 training images had no incidence of mal-registration and exhibited improved incidence of accurate registration, and higher sensitivity and specificity compared to atlases built from only one training image. Atlas-based segmentation of bone and AC from μCT images is a robust and accurate alternative to manual tissue segmentation, enabling faster, more consistent segmentation of pre-clinical datasets.

Deformation-based shape analysis of the hippocampus in the semantic variant of primary progressive aphasia and Alzheimer's disease

BACKGROUND

Increasing evidence shows that the semantic variant of primary progressive aphasia (svPPA) is characterized by hippocampal atrophy. However, less is known about disease-related morphological hippocampal changes. The goal of the present study is to conduct a detailed characterization of the impact of svPPA on global hippocampus volume and morphology compared with control subjects and patients with Alzheimer's disease (AD).

METHODS

We measured hippocampal volume and deformation-based shape differences in 22 patients with svPPA compared with 99 patients with AD and 92 controls. Multiple Automatically Generated Templates Brain Segmentation Algorithm (MAGeT-Brain) was used on MRI images obtained at the diagnostic visit.

RESULTS

Comparable left and right hippocampal atrophy were observed in svPPA and AD. Deformation-based shape analysis showed a common pattern of morphological deformation in svPPA and AD compared with controls. More specifically, both svPPA and AD showed inward deformations in the dorsal surface of the hippocampus, from head to tail on the left side, and more limited to the anterior portion of the body in the right hemisphere. These results also pointed out that both diseases are characterized by a lateral displacement of the central part (body) of the hippocampus.

DISCUSSION

Our study provides critical new evidence of hippocampal morphological changes in svPPA, similar to those found in AD. These findings highlight the importance of considering morphological hippocampal changes as part of the anatomical profile of patients with svPPA.

Automated White Matter Hyperintensity Segmentation Using Bayesian Model Selection: Assessment and Correlations with Cognitive Change

Accurate, automated white matter hyperintensity (WMH) segmentations are needed for large-scale studies to understand contributions of WMH to neurological diseases. We evaluated Bayesian Model Selection (BaMoS), a hierarchical fully-unsupervised model selection framework for WMH segmentation. We compared BaMoS segmentations to semi-automated segmentations, and assessed whether they predicted longitudinal cognitive change in control, early Mild Cognitive Impairment (EMCI), late Mild Cognitive Impairment (LMCI), subjective/significant memory concern (SMC) and Alzheimer's (AD) participants. Data were downloaded from the Alzheimer's disease Neuroimaging Initiative (ADNI). Magnetic resonance images from 30 control and 30 AD participants were selected to incorporate multiple scanners, and were semi-automatically segmented by 4 raters and BaMoS. Segmentations were assessed using volume correlation, Dice score, and other spatial metrics. Linear mixed-effect models were fitted to 180 control, 107 SMC, 320 EMCI, 171 LMCI and 151 AD participants separately in each group, with the outcomes being cognitive change (e.g. mini-mental state examination; MMSE), and BaMoS WMH, age, sex, race and education used as predictors. There was a high level of agreement between BaMoS' WMH segmentation volumes and a consensus of rater segmentations, with a median Dice score of 0.74 and correlation coefficient of 0.96. BaMoS WMH predicted cognitive change in: control, EMCI, and SMC groups using MMSE; LMCI using clinical dementia rating scale; and EMCI using Alzheimer's disease assessment scale-cognitive subscale (p < 0.05, all tests). BaMoS compares well to semi-automated segmentation, is robust to different WMH loads and scanners, and can generate volumes which predict decline. BaMoS can be applicable to further large-scale studies.