Deep Learning-Based Automatic Segmentation of Lumbosacral Nerves on CT for Spinal Intervention: A Translational Study

Early Prognostication of Critical Patients With Spinal Cord Injury: A Machine Learning Study With 1485 Cases

STUDY DESIGN

A retrospective case-series.

OBJECTIVE

The study aims to use machine learning to predict the discharge destination of spinal cord injury (SCI) patients in the intensive care unit.

SUMMARY OF BACKGROUND DATA

Prognostication following SCI is vital, especially for critical patients who need intensive care.

PATIENTS AND METHODS

Clinical data of patients diagnosed with SCI were extracted from a publicly available intensive care unit database. The first recorded data of the included patients were used to develop a total of 98 machine learning classifiers, seeking to predict discharge destination (eg, death, further medical care, home, etc.). The microaverage area under the curve (AUC) was the main indicator to assess discrimination. The best average-AUC classifier and the best death-sensitivity classifier were integrated into an ensemble classifier. The discrimination of the ensemble classifier was compared with top death-sensitivity classifiers and top average-AUC classifiers. In addition, prediction consistency and clinical utility were also assessed.

RESULTS

A total of 1485 SCI patients were included. The ensemble classifier had a microaverage AUC of 0.851, which was only slightly inferior to the best average-AUC classifier ( P =0.10). The best average-AUC classifier death sensitivity was much lower than that of the ensemble classifier. The ensemble classifier had a death sensitivity of 0.452, which was inferior to the top 8 death-sensitivity classifiers, whose microaverage AUC were inferior to the ensemble classifier ( P <0.05). In addition, the ensemble classifier demonstrated a comparable Brier score and superior net benefit in the DCA when compared with the performance of the origin classifiers.

CONCLUSIONS

The ensemble classifier shows an overall superior performance in predicting discharge destination, considering discrimination ability, prediction consistency, and clinical utility. This classifier system may aid in the clinical management of critical SCI patients in the early phase following injury.

LEVEL OF EVIDENCE

Level 3.

Automatic segmentation of dura for quantitative analysis of lumbar stenosis: A deep learning study with 518 CT myelograms

BACKGROUND

The diagnosis of lumbar spinal stenosis (LSS) can be challenging because radicular pain is not often present in the culprit-level localization. Accurate segmentation and quantitative analysis of the lumbar dura on radiographic images are key to the accurate differential diagnosis of LSS. The aim of this study is to develop an automatic dura-contouring tool for radiographic quantification on computed tomography myelogram (CTM) for patients with LSS.

METHODS

A total of 518 CTM cases with or without lumbar stenosis were included in this study. A deep learning (DL) segmentation algorithm 3-dimensional (3D) U-Net was deployed. A total of 210 labeled cases were used to develop the dura-contouring tool, with the ratio of the training, independent testing, and external validation datasets being 150:30:30. The Dice score (DCS) was the primary measure to evaluate the segmentation performance of the 3D U-Net, which was subsequently developed as the dura-contouring tool to segment another unlabeled 308 CTM cases with LSS. Automatic masks of 446 slices on the stenotic levels were then meticulously reviewed and revised by human experts, and the cross-sectional area (CSA) of the dura was compared.

RESULTS

The mean DCS of the 3D U-Net were 0.905 ± 0.080, 0.933 ± 0.018, and 0.928 ± 0.034 in the five-fold cross-validation, the independent testing, and the external validation datasets, respectively. The segmentation performance of the dura-contouring tool was also comparable to that of the second observer (the human expert). With the dura-contouring tool, only 59.0% (263/446) of the automatic masks of the stenotic slices needed to be revised. In the revised cases, there were no significant differences in the dura CSA between automatic masks and corresponding revised masks (p = 0.652). Additionally, a strong correlation of dura CSA was found between the automatic masks and corresponding revised masks (r = 0.805).

CONCLUSIONS

A dura-contouring tool was developed that could automatically segment the dural sac on CTM, and it demonstrated high accuracy and generalization ability. Additionally, the dura-contouring tool has the potential to be applied in patients with LSS because it facilitates the quantification of the dural CSA on stenotic slices.

MSEF-Net: Multi-scale edge fusion network for lumbosacral plexus segmentation with MR image

Nerve damage of spine areas is a common cause of disability and paralysis. The lumbosacral plexus segmentation from magnetic resonance imaging (MRI) scans plays an important role in many computer-aided diagnoses and surgery of spinal nerve lesions. Due to the complex structure and low contrast of the lumbosacral plexus, it is difficult to delineate the regions of edges accurately. To address this issue, we propose a Multi-Scale Edge Fusion Network (MSEF-Net) to fully enhance the edge feature in the encoder and adaptively fuse multi-scale features in the decoder. Specifically, to highlight the edge structure feature, we propose an edge feature fusion module (EFFM) by combining the Sobel operator edge detection and the edge-guided attention module (EAM), respectively. To adaptively fuse the multi-scale feature map in the decoder, we introduce an adaptive multi-scale fusion module (AMSF). Our proposed MSEF-Net method was evaluated on the collected spinal MRI dataset with 89 patients (a total of 2848 MR images). Experimental results demonstrate that our MSEF-Net is effective for lumbosacral plexus segmentation with MR images, when compared with several state-of-the-art segmentation methods.

Three-dimensional analysis of puncture needle path through safety triangle approach PLD and design of puncture positioning guide plate

OBJECTIVE

In this study, the three-dimensional relationship between the optimal puncture needle path and the lumbar spinous process was discussed using digital technology. Additionally, the positioning guide plate was designed and 3D printed in order to simulate the surgical puncture of specimens. This plate served as an important reference for the preoperative simulation and clinical application of percutaneous laser decompression (PLD).

METHOD

The CT data were imported into the Mimics program, the 3D model was rebuilt, the ideal puncture line N and the associated central axis M were developed, and the required data were measured. All of these steps were completed. A total of five adult specimens were chosen for CT scanning; the data were imported into the Mimics program; positioning guide plates were generated and 3D printed; a simulated surgical puncture of the specimens was carried out; an X-ray inspection was carried out; and an analysis of the puncture accuracy was carried out.

RESULTS

(1) The angle between line N and line M was 42°~55°, and the angles between the line M and 3D plane were 1°~2°, 5°~12°, and 78°~84°, respectively; (2) As the level of the lumbar intervertebral disc decreases, the distance from point to line and point to surface changes regularly; (3) The positioning guide was designed with the end of the lumbar spinous process and the posterior superior iliac spine on both sides as supporting points. (4) Five specimens were punctured 40 times by using the guide to simulate surgical puncture, and the success rate was 97.5%.

CONCLUSION

By analyzing the three-dimensional relationship between the optimal puncture needle path and the lumbar spinous process, the guide plate was designed to simulate surgical puncture, and the individualized safety positioning of percutaneous puncture was obtained.

Using Novel Segmentation Technology to Define Safe Corridors for Minimally Invasive Posterior Lumbar Interbody Fusion

BACKGROUND AND OBJECTIVES

There has been a rise in minimally invasive methods to access the intervertebral disk space posteriorly given their decreased tissue destruction, lower blood loss, and earlier return to work. Two such options include the percutaneous lumbar interbody fusion through the Kambin triangle and the endoscopic transfacet approach. However, without accurate preoperative visualization, these approaches carry risks of damaging surrounding structures, especially the nerve roots. Using novel segmentation technology, our goal was to analyze the anatomic borders and relative sizes of the safe triangle, trans-Kambin, and the transfacet corridors to assist surgeons in planning a safe approach and determining cannula diameters.

METHODS

The areas of the safe triangle, Kambin, and transfacet corridors were measured using commercially available software (BrainLab, Munich, Germany). For each approach, the exiting nerve root, traversing nerve roots, theca, disk, and vertebrae were manually segmented on 3-dimensional T2-SPACE magnetic resonance imaging using a region-growing algorithm. The triangles' borders were delineated ensuring no overlap between the area and the nerves.

RESULTS

A total of 11 patients (65.4 ± 12.5 years, 33.3% female) were retrospectively reviewed. The Kambin, safe, and transfacet corridors were measured bilaterally at the operative level. The mean area (124.1 ± 19.7 mm2 vs 83.0 ± 11.7 mm2 vs 49.5 ± 11.4 mm2) and maximum permissible cannula diameter (9.9 ± 0.7 mm vs 6.8 ± 0.5 mm vs 6.05 ± 0.7 mm) for the transfacet triangles were significantly larger than Kambin and the traditional safe triangles, respectively (P < .001).

CONCLUSION

We identified, in 3-dimensional, the borders for the transfacet corridor: the traversing nerve root extending inferiorly until the caudal pedicle, the theca medially, and the exiting nerve root superiorly. These results illustrate the utility of preoperatively segmenting anatomic landmarks, specifically the nerve roots, to help guide decision-making when selecting the optimal operative approach.

Machine Learning Predicts Decompression Levels for Lumbar Spinal Stenosis Using Canal Radiomic Features from Computed Tomography Myelography

BACKGROUND

The accurate preoperative identification of decompression levels is crucial for the success of surgery in patients with multi-level lumbar spinal stenosis (LSS). The objective of this study was to develop machine learning (ML) classifiers that can predict decompression levels using computed tomography myelography (CTM) data from LSS patients.

METHODS

A total of 1095 lumbar levels from 219 patients were included in this study. The bony spinal canal in CTM images was manually delineated, and radiomic features were extracted. The extracted data were randomly divided into training and testing datasets (8:2). Six feature selection methods combined with 12 ML algorithms were employed, resulting in a total of 72 ML classifiers. The main evaluation indicator for all classifiers was the area under the curve of the receiver operating characteristic (ROC-AUC), with the precision-recall AUC (PR-AUC) serving as the secondary indicator. The prediction outcome of ML classifiers was decompression level or not.

RESULTS

The embedding linear support vector (embeddingLSVC) was the optimal feature selection method. The feature importance analysis revealed the top 5 important features of the 15 radiomic predictors, which included 2 texture features, 2 first-order intensity features, and 1 shape feature. Except for shape features, these features might be eye-discernible but hardly quantified. The top two ML classifiers were embeddingLSVC combined with support vector machine (EmbeddingLSVC_SVM) and embeddingLSVC combined with gradient boosting (EmbeddingLSVC_GradientBoost). These classifiers achieved ROC-AUCs over 0.90 and PR-AUCs over 0.80 in independent testing among the 72 classifiers. Further comparisons indicated that EmbeddingLSVC_SVM appeared to be the optimal classifier, demonstrating superior discrimination ability, slight advantages in the Brier scores on the calibration curve, and Net benefits on the Decision Curve Analysis.

CONCLUSIONS

ML successfully extracted valuable and interpretable radiomic features from the spinal canal using CTM images, and accurately predicted decompression levels for LSS patients. The EmbeddingLSVC_SVM classifier has the potential to assist surgical decision making in clinical practice, as it showed high discrimination, advantageous calibration, and competitive utility in selecting decompression levels in LSS patients using canal radiomic features from CTM.

Lumbar safety triangle: comparative study of coronal and coronal oblique planes in 3.0-T magnetic resonance imaging

Objective

To compare the measurements of the lumbar safety triangle (Kambin's triangle) and the invasion of the dorsal root ganglion in the triangle in coronal and coronal oblique planes.

Materials and Methods

A cross-sectional study, in which 210 3.0-T magnetic resonance images of L2-L5 were analyzed in coronal and coronal oblique planes. Exams with lumbar spine anomalies were excluded. Demographic (sex and age) and radiological variables were recorded by a single evaluator.

Results

Most sample was female (57.1%), mean age 45.5 ± 13.3 (18-98 years). The measurements average, as well as the areas, gradually increased from L2 to L5. The dorsal root ganglion invaded the triangle in all images. The safety triangle average area was smaller in the coronal oblique plane than in the coronal plane. Of the seven dimensions of safety triangle obtained for each level of the lumbar spine, six were significantly smaller in the coronal oblique plane than in the coronal plane. The only dimension that showed no difference was the smallest ganglion dimension.

Conclusion

The dimensions and areas investigated were smaller in coronal oblique plane, especially the area (difference > 1 mm). The analysis of the triangular zone in this plane becomes important in the preoperative assessment of minimally invasive procedures.

Enhancing head and neck tumor management with artificial intelligence: Integration and perspectives

Head and neck tumors (HNTs) constitute a multifaceted ensemble of pathologies that primarily involve regions such as the oral cavity, pharynx, and nasal cavity. The intricate anatomical structure of these regions poses considerable challenges to efficacious treatment strategies. Despite the availability of myriad treatment modalities, the overall therapeutic efficacy for HNTs continues to remain subdued. In recent years, the deployment of artificial intelligence (AI) in healthcare practices has garnered noteworthy attention. AI modalities, inclusive of machine learning (ML), neural networks (NNs), and deep learning (DL), when amalgamated into the holistic management of HNTs, promise to augment the precision, safety, and efficacy of treatment regimens. The integration of AI within HNT management is intricately intertwined with domains such as medical imaging, bioinformatics, and medical robotics. This article intends to scrutinize the cutting-edge advancements and prospective applications of AI in the realm of HNTs, elucidating AI's indispensable role in prevention, diagnosis, treatment, prognostication, research, and inter-sectoral integration. The overarching objective is to stimulate scholarly discourse and invigorate insights among medical practitioners and researchers to propel further exploration, thereby facilitating superior therapeutic alternatives for patients.

Interrelations Between the Too-Long Anterior Calcaneal Process, Hind and Mid-tarsal Bone Volumes, Angles and Osteochondral Lesion of the Dome of the Talus: Analysis by Software Slicer of 69 CT Scan of Feet

Introduction

Although the association between Too-Long Anterior Calcaneal Process (TLACP) and osteochondral lesion of the dome of the talus (OCL) has been hypothesized, no study has investigated the interrelations between TLACP, hind and mid-tarsal volumes and angles and the development of OCL. The main goals of this work are: (1) to measure the volume of the calcaneum, talus, navicular and cuboid in subjects with and without TLACP; (2) to evaluate the angular relationships between talus, calcaneum and navicular in subjects with and without TLACP; (3) to assess whether TLACP has an effect on the volume of OCL.

Methods

This is a retrospective study of 69 CT scans of 54 consecutive children aged 11-15 years who had undergone a CT scan due to symptomatology suggestive of TLACP. The 3D Slicer software allowed to calculate the volume of the talus, calcaneum, navicular, cuboid, TLACP and OCL (in cm3). The PACS system was used to perform the angular measurements (in degrees) between talus, calcaneum and navicular in the frontal, axial and sagittal plane.

Results

Amid the 69 CT scans, 49 were found to have pathologies related to TLACP (71%, TLACP Group) and 20/69 were normal (29%, Control Group). The mean hind and mid-tarsal bone volumes of the TLACP group were comparable to those of the control group. There were 40 (81.6%) OCLs detected exclusively in pathological feet (TLACP group); 32 lesions were medial (80%), and 8 lesions were lateral (20%). According to Ferkel and Sgaglione CT Staging System, there were 22 (55%) stage 1 lesions, 5 (12.5%) stage 2A, 3 (7.5%) stage 2B and 10 (25%) stage three lesions. Only the angle between the talus and calcaneum in the frontal plane was significantly lower in pathological feet with respect to the control group (p < 0.001). In pathological feet, the talus was supinated, and the calcaneus pronated.

Conclusions

TLACP tend to stiffen the foot, modifying its biomechanics and leading to supination of the talus and pronation of the calcaneum. This induces an overpressure at the medial side of the talus where we observed a greater frequency of medial OCL with larger volume than lateral OCL.

Level of Evidence

III.

Three-dimensional reconstruction of Kambin's triangle based on automated magnetic resonance image segmentation

The three-dimensional (3D) anatomy of Kambin's triangle is crucial for surgical planning in minimally invasive spine surgery via the transforaminal approach. Few pieces of research have, however, used image segmentation to explore the 3D reconstruction of Kambin's triangle. This study aimed to develop a new method of 3D reconstruction of Kambin's triangle based on automated magnetic resonance image (MRI) segmentation of the lumbar spinal structures. An experienced (>5 years) "ground truth" spinal pain physician meticulously segmented and labeled spinal structures (e.g., bones, dura mater, discs, and nerve roots) on MRI. Subsequently, a 3D U-Net algorithm was developed for automatically segmenting lumbar spinal structures for the 3D reconstruction of Kambin's triangle. The Dice similarity coefficient (DSC), precision, recall, and the area of Kambin's triangle were used to assess anatomical performance. The automatic segmentation of all spinal structures at the L4/L5 levels and L5/S1 levels resulted in good performance: DSC = 0.878/0.883, precision = 0.889/0.890, recall = 0.873/0.882. Furthermore, the area measurements of Kambin's triangle revealed no significant difference between ground truth and automatic segmentation (p = 0.333 at the L4/L5 level, p = 0.302 at the L5/S1 level). The 3D U-Net model used in this study performed well in terms of simultaneous segmentation of multi-class spinal structures (including bones, dura mater, discs, and nerve roots) on MRI, allowing for accurate 3D reconstruction of Kambin's triangle.

Virtual CT Myelography: A Patch-Based Machine Learning Model to Improve Intraspinal Soft Tissue Visualization on Unenhanced Dual-Energy Lumbar Spine CT

Background: Distinguishing between the spinal cord and cerebrospinal fluid (CSF) non-invasively on CT is challenging due to their similar mass densities. We hypothesize that patch-based machine learning applied to dual-energy CT can accurately distinguish CSF from neural or other tissues based on the center voxel and neighboring voxels. Methods: 88 regions of interest (ROIs) from 12 patients’ dual-energy (100 and 140 kVp) lumbar spine CT exams were manually labeled by a neuroradiologist as one of 4 major tissue types (water, fat, bone, and nonspecific soft tissue). Four-class classifier convolutional neural networks were trained, validated, and tested on thousands of nonoverlapping patches extracted from 82 ROIs among 11 CT exams, with each patch representing pixel values (at low and high energies) of small, rectangular, 3D CT volumes. Different patch sizes were evaluated, ranging from 3 × 3 × 3 × 2 to 7 × 7 × 7 × 2. A final ensemble model incorporating all patch sizes was tested on patches extracted from six ROIs in a holdout patient. Results: Individual models showed overall test accuracies ranging from 99.8% for 3 × 3 × 3 × 2 patches (N = 19,423) to 98.1% for 7 × 7 × 7 × 2 patches (N = 1298). The final ensemble model showed 99.4% test classification accuracy, with sensitivities and specificities of 90% and 99.6%, respectively, for the water class and 98.6% and 100% for the soft tissue class. Conclusions: Convolutional neural networks utilizing local low-level features on dual-energy spine CT can yield accurate tissue classification and enhance the visualization of intraspinal neural tissue.

Artificial intelligence in spine care: current applications and future utility

PURPOSE

The field of artificial intelligence is ever growing and the applications of machine learning in spine care are continuously advancing. Given the advent of the intelligence-based spine care model, understanding the evolution of computation as it applies to diagnosis, treatment, and adverse event prediction is of great importance. Therefore, the current review sought to synthesize findings from the literature at the interface of artificial intelligence and spine research.

METHODS

A narrative review was performed based on the literature of three databases (MEDLINE, CINAHL, and Scopus) from January 2015 to March 2021 that examined historical and recent advancements in the understanding of artificial intelligence and machine learning in spine research. Studies were appraised for their role in, or description of, advancements within image recognition and predictive modeling for spinal research. Only English articles that fulfilled inclusion criteria were ultimately incorporated in this review.

RESULTS

This review briefly summarizes the history and applications of artificial intelligence and machine learning in spine. Three basic machine learning training paradigms: supervised learning, unsupervised learning, and reinforced learning are also discussed. Artificial intelligence and machine learning have been utilized in almost every facet of spine ranging from localization and segmentation techniques in spinal imaging to pathology specific algorithms which include but not limited to; preoperative risk assessment of postoperative complications, screening algorithms for patients at risk of osteoporosis and clustering analysis to identify subgroups within adolescent idiopathic scoliosis. The future of artificial intelligence and machine learning in spine surgery is also discussed with focusing on novel algorithms, data collection techniques and increased utilization of automated systems.

CONCLUSION

Improvements to modern-day computing and accessibility to various imaging modalities allow for innovative discoveries that may arise, for example, from management. Given the imminent future of AI in spine surgery, it is of great importance that practitioners continue to inform themselves regarding AI, its goals, use, and progression. In the future, it will be critical for the spine specialist to be able to discern the utility of novel AI research, particularly as it continues to pervade facets of everyday spine surgery.

Current development and prospects of deep learning in spine image analysis: a literature review

BACKGROUND AND OBJECTIVE

As the spine is pivotal in the support and protection of human bodies, much attention is given to the understanding of spinal diseases. Quick, accurate, and automatic analysis of a spine image greatly enhances the efficiency with which spine conditions can be diagnosed. Deep learning (DL) is a representative artificial intelligence technology that has made encouraging progress in the last 6 years. However, it is still difficult for clinicians and technicians to fully understand this rapidly evolving field due to the diversity of applications, network structures, and evaluation criteria. This study aimed to provide clinicians and technicians with a comprehensive understanding of the development and prospects of DL spine image analysis by reviewing published literature.

METHODS

A systematic literature search was conducted in the PubMed and Web of Science databases using the keywords "deep learning" and "spine". Date ranges used to conduct the search were from 1 January, 2015 to 20 March, 2021. A total of 79 English articles were reviewed.

KEY CONTENT AND FINDINGS

The DL technology has been applied extensively to the segmentation, detection, diagnosis, and quantitative evaluation of spine images. It uses static or dynamic image information, as well as local or non-local information. The high accuracy of analysis is comparable to that achieved manually by doctors. However, further exploration is needed in terms of data sharing, functional information, and network interpretability.

CONCLUSIONS

The DL technique is a powerful method for spine image analysis. We believe that, with the joint efforts of researchers and clinicians, intelligent, interpretable, and reliable DL spine analysis methods will be widely applied in clinical practice in the future.

A Transfer Learning Approach for Lumbar Spine Disc State Classification

Recently, deep learning algorithms have become one of the most popular methods and forms of algorithms used in the medical imaging analysis process. Deep learning tools provide accuracy and speed in the process of diagnosing and classifying lumbar spine problems. Disk herniation and spinal stenosis are two of the most common lower back diseases. The process of diagnosing pain in the lower back can be considered costly in terms of time and available expertise. In this paper, we used multiple approaches to overcome the problem of lack of training data in disc state classification and to enhance the performance of disc state classification tasks. To achieve this goal, transfer learning from different datasets and a proposed region of interest (ROI) technique were implemented. It has been demonstrated that using transfer learning from the same domain as the target dataset may increase performance dramatically. Applying the ROI method improved the disc state classification results in VGG19 2%, ResNet50 16%, MobileNetV2 5%, and VGG16 2%. The results improved VGG16 4% and in VGG19 6%, compared with the transfer from ImageNet. Moreover, it has been stated that the closer the data to be classified is to the data that the system trained on, the better the achieved results will be.

Machine Learning Algorithms in Neuroimaging: An Overview

Machine learning (ML) and artificial intelligence (AI) applications in the field of neuroimaging have been on the rise in recent years, and their clinical adoption is increasing worldwide. Deep learning (DL) is a field of ML that can be defined as a set of algorithms enabling a computer to be fed with raw data and progressively discover-through multiple layers of representation-more complex and abstract patterns in large data sets. The combination of ML and radiomics, namely the extraction of features from medical images, has proven valuable, too: Radiomic information can be used for enhanced image characterization and prognosis or outcome prediction. This chapter summarizes the basic concepts underlying ML application for neuroimaging and discusses technical aspects of the most promising algorithms, with a specific focus on Convolutional Neural Networks (CNNs) and Generative Adversarial Networks (GANs), in order to provide the readership with the fundamental theoretical tools to better understand ML in neuroimaging. Applications are highlighted from a practical standpoint in the last section of the chapter, including: image reconstruction and restoration, image synthesis and super-resolution, registration, segmentation, classification, and outcome prediction.

Surgically Relevant Morphological Parameters of the L5-S1 Interlaminar Window: A Statistical Analysis Based on 3D Reconstruction of CT Data

STUDY DESIGN

Retrospective study.

OBJECTIVES

The interlaminar window is the most important anatomical corridor for posterior approach of lumbar procedures. Three-dimensional (3D) reconstruction of the L5-S1 interlaminar window may benefit the accurate measurement and assessment of surgical considerations. The aim of this study was to measure surgical relevant parameters of the L5-S1 interlaminar window based on 3D reconstruction of lumbar CTs.

METHODS

50 thin-layer CT data were retrospectively collected, segmented, and reconstructed. Surgical relevant parameters included the width, left height, right height, interpedicular distance, area, and operable area of the L5-S1 interlaminar window. Morphological measurements were performed independently by two experienced experts. Patients with radiologic abnormalities at L5-S1 level were regarded as group A (n=28), while those without L5-S1 disc herniation were regarded as group B (n=22).

RESULTS

The average left height, right height, width, and area of the L5-S1 interlaminar window were 9.14±2.45mm, 9.55±2.56mm, 23.55±4.91mm, and 144.57±57.05mm2. The average interpedicular distance (IPD) at superior, middle, and inferior pedicle level were 29.29±3.39mm, 27.96±3.38mm and 37.46±4.23mm, with significant differences among these three parameters (P<0.05). The average operable areas of the L5-S1 interlaminar window were: left-axilla 24.52±15.91mm2, left-shoulder 27.14±15.48mm2, right-axilla 29.95±17.17mm2, and right-shoulder 31.12±16.40mm2 (P>0.05). There were no significant differences between group A and B in these parameters (P>0.05), except the inferior IPD (36.69±3.73mm vs 39.23±3.01mm, P=0.017<0.05).

CONCLUSION

The morphological measurement of the L5-S1 interlaminar window based on 3D reconstruction provided accurate and reliable reference data for epidural puncture approach and posterior approach of lumbar surgery. Moreover, it could also assist the placement of endoscopic working channel in percutaneous endoscopic interlaminar discectomy (PEID) and might be useful for further studies of anatomical and surgical consideration of unilateral biportal endoscopic spinal surgery (UBE) procedures. Key words: Interlaminar window; percutaneous endoscopic interlaminar discectomy (PEID); unilateral biportal endoscopy spinal surgery (UBE); 3D reconstruction.

Clinical Artificial Intelligence Applications in Radiology: Neuro

Radiologists have been at the forefront of the digitization process in medicine. Artificial intelligence (AI) is a promising area of innovation, particularly in medical imaging. The number of applications of AI in neuroradiology has also grown. This article illustrates some of these applications. This article reviews machine learning challenges related to neuroradiology. The first approval of reimbursement for an AI algorithm by the Centers for Medicare and Medicaid Services, covering a stroke software for early detection of large vessel occlusion, is also discussed.

Impact of transfer learning for human sperm segmentation using deep learning

BACKGROUND AND OBJECTIVE

Infertility affects approximately one in ten couples, and almost half of the infertility cases are due to the malefactor. To diagnose infertility and determine future treatment, a semen analysis is performed. Evaluation of sperm morphology is one of several steps in semen analysis, in which the shape and size of sperm parts are examined. The laboratories dedicated to this use traditional methods susceptible to errors. An alternative to replace the poor visual ability to assess sperm size and shape is to analyze sperm morphology with a computer's help. However, since the automatic sperm classification rates do not show an acceptable precision rate for use in the clinical setting, it is considered an exciting approach to focus efforts on improving the precision in sperm segmentation to extract the contour sperm before classification. This work aims to assess the utility of two image segmentation deep learning models for segmenting human sperm heads, acrosome, and nucleus.

METHODS

In this work, we evaluate the use of two well-known deep learning architectures (U-Net and Mask-RCNN) to segment parts of human sperm cells using data augmentation, cross-validation, hyperparameter tuning, and transfer learning. The experimental results are carried out using SCIAN-SpermSegGS, a public dataset with more than two hundred manually segmented sperm cells and widely used to validate segmentation methods of human sperm parts.

RESULTS

Experimental evaluation shows that U-net with transfer learning achieves up to 95% overlapping against hand-segmented masks for sperm head (0.96), acrosome (0.94), and nucleus (0.95), using Dice coefficient as the evaluation metric. These results outperform state-of-the-art sperm parts segmentation methods.

CONCLUSIONS

The impact of transfer learning is substantial, significantly improving the results of state-of-the-art methods with a higher Dice coefficient, less dispersion, and fewer cases where the model failed to segment sperm parts. These results represent a promising advance in the ultimate goal of performing computer-assisted morphological sperm analysis.

A novel dual-network architecture for mixed-supervised medical image segmentation

In medical image segmentation tasks, deep learning-based models usually require densely and precisely annotated datasets to train, which are time-consuming and expensive to prepare. One possible solution is to train with the mixed-supervised dataset, where only a part of data is densely annotated with segmentation map and the rest is annotated with some weak form, such as bounding box. In this paper, we propose a novel network architecture called Mixed-Supervised Dual-Network (MSDN), which consists of two separate networks for the segmentation and detection tasks respectively, and a series of connection modules between the layers of the two networks. These connection modules are used to extract and transfer useful information from the detection task to help the segmentation task. We exploit a variant of a recently designed technique called 'Squeeze and Excitation' in the connection module to boost the information transfer between the two tasks. Compared with existing model with shared backbone and multiple branches, our model has flexible and trainable feature sharing fashion and thus is more effective and stable. We conduct experiments on 4 medical image segmentation datasets, and experiment results show that the proposed MSDN model outperforms multiple baselines.

Artificial Intelligence-enabled, Real-time Intraoperative Ultrasound Imaging of Neural Structures Within the Psoas: Validation in a Porcine Spine Model

STUDY DESIGN

Experimental in-vivo animal study.

OBJECTIVE

The aim of this study was to evaluate an Artificial Intelligence (AI)-enabled ultrasound imaging system's ability to detect, segment, classify, and display neural and other structures during trans-psoas spine surgery.

SUMMARY OF BACKGROUND DATA

Current methodologies for intraoperatively localizing and visualizing neural structures within the psoas are limited and can impact the safety of lateral lumbar interbody fusion (LLIF). Ultrasound technology, enhanced with AI-derived neural detection algorithms, could prove useful for this task.

METHODS

The study was conducted using an in vivo porcine model (50 subjects). Image processing and machine learning algorithms were developed to detect neural and other anatomic structures within and adjacent to the psoas muscle while using an ultrasound imaging system during lateral lumbar spine surgery (SonoVision,™ Tissue Differentiation Intelligence, USA). The imaging system's ability to detect and classify the anatomic structures was assessed with subsequent tissue dissection. Dice coefficients were calculated to quantify the performance of the image segmentation.

RESULTS

The AI-trained ultrasound system detected, segmented, classified, and displayed nerve, psoas muscle, and vertebral body surface with high sensitivity and specificity. The mean Dice coefficient score for each tissue type was >80%, indicating that the detected region and ground truth were >80% similar to each other. The mean specificity of nerve detection was 92%; for bone and muscle, it was >95%. The accuracy of nerve detection was >95%.

CONCLUSION

This study demonstrates that a combination of AI-derived image processing and machine learning algorithms can be developed to enable real-time ultrasonic detection, segmentation, classification, and display of critical anatomic structures, including neural tissue, during spine surgery. AI-enhanced ultrasound imaging can provide a visual map of important anatomy in and adjacent to the psoas, thereby providing the surgeon with critical information intended to increase the safety of LLIF surgery.Level of Evidence: N/A.

Deep learning-based lumbosacral reconstruction for difficulty prediction of percutaneous endoscopic transforaminal discectomy at L5/S1 level: A retrospective cohort study

BACKGROUND

Deep learning has been validated as a promising technique for automatic segmentation and rapid three-dimensional (3D) reconstruction of lumbosacral structures on CT. Simulated foraminoplasty of percutaneous endoscopic transforaminal discectomy (PETD) through the Kambin triangle may benefit viability assessment of PETD at L5/S1 level.

MATERIAL AND METHODS

Medical records and radiographic data of patients with L5/S1 lumbar disc herniation (LDH) who received a single-level PETD from March 2013 to February 2018 were retrospectively collected and analyzed. Deep learning was adopted to achieve semantic segmentation of lumbosacral structures (nerve, bone, disc) on CT, and the segmented masks on reconstructed 3D models. Two observers measured the area of the Kambin triangle on 6 selected deep learning-derived 3D (DL-3D) models and ground truth-derived 3D (GT-3D) models, and intraclass correlation coefficient (ICC) was calculated to assess the test-retest and interobserver reliability. Foraminoplasty of PETD was simulated on L5/S1 lumbosacral 3D models. Patients with extended foraminoplasty or stuck canula occurs on simulations were predicted as PETD-difficult cases (Group A). The remaining patients were regarded as PETD-normal cases (Group B). Clinical information and outcomes were compared between the two groups.

RESULTS

Deep learning-derived 3D models of lumbosacral structures (nerves, bones, and disc) from thin-layer CT were reliable. The area of the Kambin triangle was 161.27 ± 40.10 mm² on DL-3D models and 153.57 ± 32.37 mm² on GT-3D models (p = 0.206). Reliability test revealed strong test-retest reliability (ICC between 0.947 and 0.971) and interobserver reliability of multiple measurements (ICC between 0.866 and 0.961). The average operation time was 99.62 ± 17.39 min in Group A and 88.93 ± 21.87 min in Group B (P = 0.025). No significant differences in patient-reported outcomes or complications were observed between the two groups (P > 0.05).

CONCLUSION

Deep learning achieved accurate and rapid segmentations of lumbosacral structures on CT, and deep learning-based 3D reconstructions were efficacious and reliable. Foraminoplasty simulation with deep learning-based lumbosacral reconstructions may benefit surgical difficulty prediction of PETD at L5/S1 level.

Automatic Vertebral Body Segmentation Based on Deep Learning of Dixon Images for Bone Marrow Fat Fraction Quantification

Background: Bone marrow fat (BMF) fraction quantification in vertebral bodies is used as a novel imaging biomarker to assess and characterize chronic lower back pain. However, manual segmentation of vertebral bodies is time consuming and laborious. Purpose: (1) Develop a deep learning pipeline for segmentation of vertebral bodies using quantitative water-fat MRI. (2) Compare BMF measurements between manual and automatic segmentation methods to assess performance. Materials and Methods: In this retrospective study, MR images using a 3D spoiled gradient-recalled echo (SPGR) sequence with Iterative Decomposition of water and fat with Echo Asymmetry and Least-squares estimation (IDEAL) reconstruction algorithm were obtained in 57 subjects (28 women, 29 men, mean age, 47.2 ± 12.6 years). An artificial network was trained for 100 epochs on a total of 165 lumbar vertebrae manually segmented from 31 subjects. Performance was assessed by analyzing the receiver operating characteristic curve, precision-recall, F1 scores, specificity, sensitivity, and similarity metrics. Bland-Altman analysis was used to assess performance of BMF fraction quantification using the predicted segmentations. Results: The deep learning segmentation method achieved an AUC of 0.92 (CI 95%: 0.9186, 0.9195) on a testing dataset (n = 24 subjects) on classification of pixels as vertebrae. A sensitivity of 0.99 and specificity of 0.80 were achieved for a testing dataset, and a mean Dice similarity coefficient of 0.849 ± 0.091. Comparing manual and automatic segmentations on fat fraction maps of lumbar vertebrae (n = 124 vertebral bodies) using Bland-Altman analysis resulted in a bias of only -0.605% (CI 95% = -0.847 to -0.363%) and agreement limits of -3.275% and +2.065%. Automatic segmentation was also feasible in 16 ± 1 s. Conclusion: Our results have demonstrated the feasibility of automated segmentation of vertebral bodies using deep learning models on water-fat MR (Dixon) images to define vertebral regions of interest with high specificity. These regions of interest can then be used to quantify BMF with comparable results as manual segmentation, providing a framework for completely automated investigation of vertebral changes in CLBP.

Artificial Intelligence in Head and Neck Imaging: A Glimpse into the Future

Artificial intelligence, specifically machine learning and deep learning, is a rapidly developing field in imaging sciences with the potential to improve the efficiency and effectiveness of radiologists. This review covers common technical terms and basic concepts in imaging artificial intelligence and briefly reviews the application of these techniques to general imaging as well as head and neck imaging. Artificial intelligence has the potential to contribute improvements to all areas of patient care, including image acquisition, processing, segmentation, automated detection of findings, integration of clinical information, quality improvement, and research. Numerous challenges remain, however, before widespread imaging clinical adoption and integration occur.

Mixed-Supervised Dual-Network for Medical Image Segmentation

Deep learning based medical image segmentation models usually require large datasets with high-quality dense segmentations to train, which are very time-consuming and expensive to prepare. One way to tackle this difficulty is using the mixed-supervised learning framework, where only a part of data is densely annotated with segmentation label and the rest is weakly labeled with bounding boxes. The model is trained jointly in a multi-task learning setting. In this paper, we propose Mixed-Supervised Dual-Network (MSDN), a novel architecture which consists of two separate networks for the detection and segmentation tasks respectively, and a series of connection modules between the layers of the two networks. These connection modules are used to transfer useful information from the auxiliary detection task to help the segmentation task. We propose to use a recent technique called 'Squeeze and Excitation' in the connection module to boost the transfer. We conduct experiments on two medical image segmentation datasets. The proposed MSDN model outperforms multiple baselines.