Harmonization Strategies in Multicenter MRI-Based Radiomics

Deep feature batch correction using ComBat for machine learning applications in computational pathology

Background

Developing artificial intelligence (AI) models for digital pathology requires large datasets from multiple sources. However, without careful implementation, AI models risk learning confounding site-specific features in datasets instead of clinically relevant information, leading to overestimated performance, poor generalizability to real-world data, and potential misdiagnosis.

Methods

Whole-slide images (WSIs) from The Cancer Genome Atlas (TCGA) colon (COAD), and stomach adenocarcinoma datasets were selected for inclusion in this study. Patch embeddings were obtained using three feature extraction models, followed by ComBat harmonization. Attention-based multiple instance learning models were trained to predict tissue-source site (TSS), as well as clinical and genetic attributes, using raw, Macenko normalized, and Combat-harmonized patch embeddings.

Results

TSS prediction achieved high accuracy (AUROC > 0.95) with all three feature extraction models. ComBat harmonization significantly reduced the AUROC for TSS prediction, with mean AUROCs dropping to approximately 0.5 for most models, indicating successful mitigation of batch effects (e.g., CCL-ResNet50 in TCGA-COAD: Pre-ComBat AUROC = 0.960, Post-ComBat AUROC = 0.506, p < 0.001). Clinical attributes associated with TSS, such as race and treatment response, showed decreased predictability post-harmonization. Notably, the prediction of genetic features like MSI status remained robust after harmonization (e.g., MSI in TCGA-COAD: Pre-ComBat AUROC = 0.667, Post-ComBat AUROC = 0.669, p=0.952), indicating the preservation of true histological signals.

Conclusion

ComBat harmonization of deep learning-derived histology features effectively reduces the risk of AI models learning confounding features in WSIs, ensuring more reliable performance estimates. This approach is promising for the integration of large-scale digital pathology datasets.

Radiomic study of antenatal prediction of severe placenta accreta spectrum from MRI

OBJECTIVES

We previously demonstrated the potential of radiomics for the prediction of severe histological placenta accreta spectrum (PAS) subtypes using T2-weighted MRI. We aim to validate our model using an additional dataset. Secondly, we explore whether the performance is improved using a new approach to develop a new multivariate radiomics model.

METHODS

Multi-centre retrospective analysis was conducted between 2018 and 2023. Inclusion criteria: MRI performed for suspicion of PAS from ultrasound, clinical findings of PAS at laparotomy and/or histopathological confirmation. Radiomic features were extracted from T2-weighted MRI. The previous multivariate model was validated. Secondly, a 5-radiomic feature random forest classifier was selected from a randomized feature selection scheme to predict invasive placenta increta PAS cases. Prediction performance was assessed based on several metrics including area under the curve (AUC) of the receiver operating characteristic curve (ROC), sensitivity, and specificity.

RESULTS

We present 100 women [mean age 34.6 (±3.9) with PAS], 64 of whom had placenta increta. Firstly, we validated the previous multivariate model and found that a support vector machine classifier had a sensitivity of 0.620 (95% CI: 0.068; 1.0), specificity of 0.619 (95% CI: 0.059; 1.0), an AUC of 0.671 (95% CI: 0.440; 0.922), and accuracy of 0.602 (95% CI: 0.353; 0.817) for predicting placenta increta. From the new multivariate model, the best 5-feature subset was selected via the random subset feature selection scheme comprised of 4 radiomic features and 1 clinical variable (number of previous caesareans). This clinical-radiomic model achieved an AUC of 0.713 (95% CI: 0.551; 0.854), accuracy of 0.695 (95% CI 0.563; 0.793), sensitivity of 0.843 (95% CI 0.682; 0.990), and specificity of 0.447 (95% CI 0.167; 0.667).

CONCLUSION

We validated our previous model and present a new multivariate radiomic model for the prediction of severe placenta increta from a well-defined, cohort of PAS cases.

ADVANCES IN KNOWLEDGE

Radiomic features demonstrate good predictive potential for identifying placenta increta. This suggests radiomics may be a useful adjunct to clinicians caring for women with this high-risk pregnancy condition.

ESR Essentials: radiomics-practice recommendations by the European Society of Medical Imaging Informatics

Radiomics is a method to extract detailed information from diagnostic images that cannot be perceived by the naked eye. Although radiomics research carries great potential to improve clinical decision-making, its inherent methodological complexities make it difficult to comprehend every step of the analysis, often causing reproducibility and generalizability issues that hinder clinical adoption. Critical steps in the radiomics analysis and model development pipeline-such as image, application of image filters, and selection of feature extraction parameters-can greatly affect the values of radiomic features. Moreover, common errors in data partitioning, model comparison, fine-tuning, assessment, and calibration can reduce reproducibility and impede clinical translation. Clinical adoption of radiomics also requires a deep understanding of model explainability and the development of intuitive interpretations of radiomic features. To address these challenges, it is essential for radiomics model developers and clinicians to be well-versed in current best practices. Proper knowledge and application of these practices is crucial for accurate radiomics feature extraction, robust model development, and thorough assessment, ultimately increasing reproducibility, generalizability, and the likelihood of successful clinical translation. In this article, we have provided researchers with our recommendations along with practical examples to facilitate good research practices in radiomics. KEY POINTS: Radiomics' inherent methodological complexity should be understood to ensure rigorous radiomic model development to improve clinical decision-making. Adherence to radiomics-specific checklists and quality assessment tools ensures methodological rigor. Use of standardized radiomics tools and best practices enhances clinical translation of radiomics models.

The Pattern and Staging of Brain Atrophy in Spinocerebellar Ataxia Type 2 (SCA2): MRI Volumetrics from ENIGMA-Ataxia

Objective

Spinocerebellar ataxia type 2 (SCA2) is a rare, inherited neurodegenerative disease characterised by progressive deterioration in both motor coordination and cognitive function. Atrophy of the cerebellum, brainstem, and spinal cord are core features of SCA2, however the evolution and pattern of whole-brain atrophy in SCA2 remain unclear. We undertook a multi-site, structural magnetic resonance imaging (MRI) study to comprehensively characterize the neurodegeneration profile of SCA2.

Methods

Voxel-based morphometry analyses of 110 participants with SCA2 and 128 controls were undertaken to assess groupwise differences in whole-brain volume. Correlations with clinical severity and genotype, and cross-sectional profiling of atrophy patterns at different disease stages, were also performed.

Results

Atrophy in SCA2 relative to controls was greatest (Cohen's d>2.5) in the cerebellar white matter (WM), middle cerebellar peduncle, pons, and corticospinal tract. Very large effects (d>1.5) were also evident in the superior cerebellar, inferior cerebellar, and cerebral peduncles. In cerebellar grey matter (GM), large effects (d>0.8) mapped to areas related to both motor coordination and cognitive tasks. Strong correlations (|r|>0.4) between volume and disease severity largely mirrored these groupwise outcomes. Stratification by disease severity showed a degeneration pattern beginning in cerebellar and pontine WM in pre-clinical subjects; spreading to the cerebellar GM and cerebro-cerebellar/corticospinal WM tracts; then finally involving the thalamus, striatum, and cortex in severe stages.

Interpretation

The magnitude and pattern of brain atrophy evolves over the course of SCA2, with widespread, non-uniform involvement across the brainstem, cerebellar tracts, and cerebellar cortex; and late involvement of the cerebral cortex and striatum.

Deep learning for the harmonization of structural MRI scans: a survey

Medical imaging datasets for research are frequently collected from multiple imaging centers using different scanners, protocols, and settings. These variations affect data consistency and compatibility across different sources. Image harmonization is a critical step to mitigate the effects of factors like inherent differences between various vendors, hardware upgrades, protocol changes, and scanner calibration drift, as well as to ensure consistent data for medical image processing techniques. Given the critical importance and widespread relevance of this issue, a vast array of image harmonization methodologies have emerged, with deep learning-based approaches driving substantial advancements in recent times. The goal of this review paper is to examine the latest deep learning techniques employed for image harmonization by analyzing cutting-edge architectural approaches in the field of medical image harmonization, evaluating both their strengths and limitations. This paper begins by providing a comprehensive fundamental overview of image harmonization strategies, covering three critical aspects: established imaging datasets, commonly used evaluation metrics, and characteristics of different scanners. Subsequently, this paper analyzes recent structural MRI (Magnetic Resonance Imaging) harmonization techniques based on network architecture, network learning algorithm, network supervision strategy, and network output. The underlying architectures include U-Net, Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), flow-based generative models, transformer-based approaches, as well as custom-designed network architectures. This paper investigates the effectiveness of Disentangled Representation Learning (DRL) as a pivotal learning algorithm in harmonization. Lastly, the review highlights the primary limitations in harmonization techniques, specifically the lack of comprehensive quantitative comparisons across different methods. The overall aim of this review is to serve as a guide for researchers and practitioners to select appropriate architectures based on their specific conditions and requirements. It also aims to foster discussions around ongoing challenges in the field and shed light on promising future research directions with the potential for significant advancements.

Deep Learning-Driven Transformation: A Novel Approach for Mitigating Batch Effects in Diffusion MRI Beyond Traditional Harmonization

BACKGROUND

"Batch effect" in MR images, due to vendor-specific features, MR machine generations, and imaging parameters, challenges image quality and hinders deep learning (DL) model generalizability.

PURPOSE

We aim to develop a DL model using contrast adjustment and super-resolution to reduce diffusion-weighted images (DWIs) diversity across magnetic field strengths and imaging parameters.

STUDY TYPE

Retrospective.

SUBJECTS

The DL model was built using an open dataset from one individual. The MR machine identification model was trained and validated on a dataset of 1134 adults (54% females, 46% males), with 1050 subjects showing no DWI abnormalities and 84 with conditions like stroke and tumors. The 21,000 images were divided into 80% for training, 20% for validation, and 3500 for testing.

FIELD STRENGTH/SEQUENCE

Seven MR scanners from four manufacturers with 1.5 T and 3 T magnetic field strengths. DWIs were acquired using spin-echo sequences and high-resolution T2WIs using the T2-SPACE sequence.

ASSESSMENT

An experienced, board-certified radiologist evaluated the effectiveness of restoring high-resolution T2WI and harmonizing diverse DWI with metrics such as PSNR and SSIM, and the texture and frequency attributes were further analyzed using gray-level co-occurrence matrix and 1-dimensional power spectral density. The model's impact on machine-specific characteristics was gauged through the performance metrics of a ResNet-50 model. Comprehensive statistical tests were employed for statistical robustness, including McNemar's test and the Dice index.

RESULTS

Our DL protocol reduced DWI contrast and resolution variation. ResNet-50 model's accuracy decreased from 0.9443 to 0.5786, precision from 0.9442 to 0.6494, recall from 0.9443 to 0.5786, and F1 score from 0.9438 to 0.5587. The t-SNE visualization indicated more consistent image features across multiple MR devices. Autoencoder halved learning iterations; Dice coefficient >0.74 confirmed signal reproducibility in 84 lesions.

CONCLUSION

This study presents a DL strategy to mitigate batch effects in diffusion MR images, improving their quality and generalizability.

EVIDENCE LEVEL

3 TECHNICAL EFFICACY: Stage 1.

Mitigating the impact of image processing variations on tumour [18F]-FDG-PET radiomic feature robustness

Radiomics analysis of [18F]-fluorodeoxyglucose ([18F]-FDG) PET images could be leveraged for personalised cancer medicine. However, the inherent sensitivity of radiomic features to intensity discretisation and voxel interpolation complicates its clinical translation. In this work, we evaluated the robustness of tumour [18F]-FDG-PET radiomic features to 174 different variations in intensity resolution or voxel size, and determined whether implementing parameter range conditions or dependency corrections could improve their robustness. Using 485 patient images spanning three cancer types: non-small cell lung cancer (NSCLC), melanoma, and lymphoma, we observed features were more sensitive to intensity discretisation than voxel interpolation, especially texture features. In most of our investigations, the majority of non-robust features could be made robust by applying parameter range conditions. Correctable features, which were generally fewer than conditionally robust, showed systematic dependence on bin configuration or voxel size that could be minimised by applying corrections based on simple mathematical equations. Melanoma images exhibited limited robustness and correctability relative to NSCLC and lymphoma. Our study provides an in-depth characterisation of the sensitivity of [18F]-FDG-PET features to image processing variations and reinforces the need for careful selection of imaging biomarkers prior to any clinical application.

Automated glioblastoma patient classification using hypoxia levels measured through magnetic resonance images

INTRODUCTION

The challenge of treating Glioblastoma (GBM) tumors is due to various mechanisms that make the tumor resistant to radiation therapy. One of these mechanisms is hypoxia, and therefore, determining the level of hypoxia can improve treatment planning and initial evaluation of its effectiveness in GBM. This study aimed to design an intelligent system to classify glioblastoma patients based on hypoxia levels obtained from magnetic resonance images with the help of an artificial neural network (ANN).

MATERIAL AND METHOD

MR images and PET measurements were available for this study. MR images were downloaded from the Cancer Imaging Archive (TCIA) database to classify glioblastoma patients based on hypoxia. The images in this database were prepared from 27 patients with glioblastoma on T1W + Gd, T2W-FLAIR, and T2W. Our designed algorithm includes various parts of pre-processing, tumor segmentation, feature extraction from images, and matching these features with quantitative parameters related to hypoxia in PET images. The system's performance is evaluated by categorizing glioblastoma patients based on hypoxia.

RESULTS

The results of classification with the artificial neural network (ANN) algorithm were as follows: the highest sensitivity, specificity, and accuracy were obtained at 86.71, 85.99 and 83.17%, respectively. The best specificity was related to the T2W-EDEMA image with the tumor to blood ratio (TBR) as a hypoxia parameter. T1W-NECROSIS image with the TBR parameter also showed the highest sensitivity and accuracy.

CONCLUSION

The results of the present study can be used in clinical procedures before treating glioblastoma patients. Among these treatment approaches, we can mention the radiotherapy treatment design and the prescription of effective drugs for the treatment of hypoxic tumors.

Down-sampling in diffusion MRI: a bundle-specific DTI and NODDI study

Introduction

Multi-shell diffusion Magnetic Resonance Imaging (dMRI) data has been widely used to characterise white matter microstructure in several neurodegenerative diseases. The lack of standardised dMRI protocols often implies the acquisition of redundant measurements, resulting in prolonged acquisition times. In this study, we investigate the impact of the number of gradient directions on Diffusion Tensor Imaging (DTI) and on Neurite Orientation Dispersion and Density Imaging (NODDI) metrics.

Methods

Data from 124 healthy controls collected in three different longitudinal studies were included. Using an in-house algorithm, we reduced the number of gradient directions in each data shell. We estimated DTI and NODDI measures on six white matter bundles clinically relevant for neurodegenerative diseases.

Results

Fractional Anisotropy (FA) measures on bundles where data were sampled at the 30% rate, showed a median L1 distance of up to 3.92% and a 95% CI of (1.74, 8.97)% when compared to those obtained at reference sampling. Mean Diffusivity (MD) reached up to 4.31% and a 95% CI of (1.60, 16.98)% on the same premises. At a sampling rate of 50%, we obtained a median of 3.90% and a 95% CI of (1.99, 16.65)% in FA, and 5.49% with a 95% CI of (2.14, 21.68)% in MD. The Intra-Cellular volume fraction (ICvf) median L1 distance was up to 2.83% with a 95% CI of (1.98, 4.82)% at a 30% sampling rate and 3.95% with a 95% CI of (2.39, 7.81)% at a 50% sampling rate. The volume difference of the reconstructed white matter at reference and 50% sampling reached a maximum of (2.09 ± 0.81)%.

Discussion

In conclusion, DTI and NODDI measures reported at reference sampling were comparable to those obtained when the number of dMRI volumes was reduced by up to 30%. Close to reference DTI and NODDI metrics were estimated with a significant reduction in acquisition time using three shells, respectively with: 4 directions at a b value of 700 s/mm2, 14 at 1000 s/mm2, and 32 at 2000 s/mm2. The study revealed aspects that can be important for large-scale clinical studies on bundle-specific diffusion MRI.

Incorporating a-priori information in deep learning models for quantitative susceptibility mapping via adaptive convolution

Quantitative susceptibility mapping (QSM) has attracted considerable interest for tissue characterization (e.g., iron and calcium accumulation, myelination, venous vasculature) in the human brain and relies on extensive data processing of gradient-echo MRI phase images. While deep learning-based field-to-susceptibility inversion has shown great potential, the acquisition parameters applied in clinical settings such as image resolution or image orientation with respect to the magnetic field have not been fully accounted for. Furthermore, the lack of comprehensive training data covering a wide range of acquisition parameters further limits the current QSM deep learning approaches. Here, we propose the integration of a priori information of imaging parameters into convolutional neural networks with our approach, adaptive convolution, that learns the mapping between the additional presented information (acquisition parameters) and the changes in the phase images associated with these varying acquisition parameters. By associating a-priori information with the network parameters itself, the optimal set of convolution weights is selected based on data-specific attributes, leading to generalizability towards changes in acquisition parameters. Moreover, we demonstrate the feasibility of pre-training on synthetic data and transfer learning to clinical brain data to achieve substantial improvements in the computation of susceptibility maps. The adaptive convolution 3D U-Net demonstrated generalizability in acquisition parameters on synthetic and in-vivo data and outperformed models lacking adaptive convolution or transfer learning. Further experiments demonstrate the impact of the side information on the adaptive model and assessed susceptibility map computation on simulated pathologic data sets and measured phase data.

Simulation of acquisition shifts in T2 weighted fluid-attenuated inversion recovery magnetic resonance images to stress test artificial intelligence segmentation networks

Purpose

To provide a simulation framework for routine neuroimaging test data, which allows for "stress testing" of deep segmentation networks against acquisition shifts that commonly occur in clinical practice for T2 weighted (T2w) fluid-attenuated inversion recovery magnetic resonance imaging protocols.

Approach

The approach simulates "acquisition shift derivatives" of MR images based on MR signal equations. Experiments comprise the validation of the simulated images by real MR scans and example stress tests on state-of-the-art multiple sclerosis lesion segmentation networks to explore a generic model function to describe the F1 score in dependence of the contrast-affecting sequence parameters echo time (TE) and inversion time (TI).

Results

The differences between real and simulated images range up to 19% in gray and white matter for extreme parameter settings. For the segmentation networks under test, the F1 score dependency on TE and TI can be well described by quadratic model functions (R 2 > 0.9 ). The coefficients of the model functions indicate that changes of TE have more influence on the model performance than TI.

Conclusions

We show that these deviations are in the range of values as may be caused by erroneous or individual differences in relaxation times as described by literature. The coefficients of the F1 model function allow for a quantitative comparison of the influences of TE and TI. Limitations arise mainly from tissues with a low baseline signal (like cerebrospinal fluid) and when the protocol contains contrast-affecting measures that cannot be modeled due to missing information in the DICOM header.

Early prognostication of overall survival for pediatric diffuse midline gliomas using MRI radiomics and machine learning: a two-center study

Background

Diffuse midline gliomas (DMG) are aggressive pediatric brain tumors that are diagnosed and monitored through MRI. We developed an automatic pipeline to segment subregions of DMG and select radiomic features that predict patient overall survival (OS).

Methods

We acquired diagnostic and post-radiation therapy (RT) multisequence MRI (T1, T1ce, T2, T2 FLAIR) and manual segmentations from two centers of 53 (internal cohort) and 16 (external cohort) DMG patients. We pretrained a deep learning model on a public adult brain tumor dataset, and finetuned it to automatically segment tumor core (TC) and whole tumor (WT) volumes. PyRadiomics and sequential feature selection were used for feature extraction and selection based on the segmented volumes. Two machine learning models were trained on our internal cohort to predict patient 1-year survival from diagnosis. One model used only diagnostic tumor features and the other used both diagnostic and post-RT features.

Results

For segmentation, Dice score (mean [median]±SD) was 0.91 (0.94)±0.12 and 0.74 (0.83)±0.32 for TC, and 0.88 (0.91)±0.07 and 0.86 (0.89)±0.06 for WT for internal and external cohorts, respectively. For OS prediction, accuracy was 77% and 81% at time of diagnosis, and 85% and 78% post-RT for internal and external cohorts, respectively. Homogeneous WT intensity in baseline T2 FLAIR and larger post-RT TC/WT volume ratio indicate shorter OS.

Conclusions

Machine learning analysis of MRI radiomics has potential to accurately and non-invasively predict which pediatric patients with DMG will survive less than one year from the time of diagnosis to provide patient stratification and guide therapy.

Early prognostication of overall survival for pediatric diffuse midline gliomas using MRI radiomics and machine learning: A two-center study

Background

Diffuse midline gliomas (DMG) are aggressive pediatric brain tumors that are diagnosed and monitored through MRI. We developed an automatic pipeline to segment subregions of DMG and select radiomic features that predict patient overall survival (OS).

Methods

We acquired diagnostic and post-radiation therapy (RT) multisequence MRI (T1, T1ce, T2, and T2 FLAIR) and manual segmentations from 2 centers: 53 from 1 center formed the internal cohort and 16 from the other center formed the external cohort. We pretrained a deep learning model on a public adult brain tumor data set (BraTS 2021), and finetuned it to automatically segment tumor core (TC) and whole tumor (WT) volumes. PyRadiomics and sequential feature selection were used for feature extraction and selection based on the segmented volumes. Two machine learning models were trained on our internal cohort to predict patient 12-month survival from diagnosis. One model used only data obtained at diagnosis prior to any therapy (baseline study) and the other used data at both diagnosis and post-RT (post-RT study).

Results

Overall survival prediction accuracy was 77% and 81% for the baseline study, and 85% and 78% for the post-RT study, for internal and external cohorts, respectively. Homogeneous WT intensity in baseline T2 FLAIR and larger post-RT TC/WT volume ratio indicate shorter OS.

Conclusions

Machine learning analysis of MRI radiomics has potential to accurately and noninvasively predict which pediatric patients with DMG will survive less than 12 months from the time of diagnosis to provide patient stratification and guide therapy.

Evaluation of calcification distribution by CT-based textural analysis for discrimination of immature teratoma

BACKGROUND

Mature and immature teratomas are differentiated based on tumor markers and calcification or fat distribution. However, no study has objectively quantified the differences in calcification and fat distributions between these tumors. This study aimed to evaluate the diagnostic potential of CT-based textural analysis in differentiating between mature and immature teratomas in patients aged < 20 years.

MATERIALS AND METHODS

Thirty-two patients with pathologically proven mature cystic (n = 28) and immature teratomas (n = 4) underwent transabdominal ultrasound and/or abdominal and pelvic CT before surgery. The diagnostic performance of CT for assessing imaging features, including subjective manual measurement and objective textural analysis of fat and calcification distributions in the tumors, was evaluated by two experienced readers. The histopathological results were used as the gold standard. The Mann-Whitney U test was used for statistical analysis.

RESULTS

We evaluated 32 patients (mean age, 14.5 years; age range, 6-19 years). The mean maximum diameter and number of calcifications of immature teratomas were significantly larger than those of mature cystic teratomas (p < 0.01). The mean number of fats of immature teratomas was significantly larger than that of mature cystic teratomas (p < 0.01); however, no significant difference in the maximum diameter of fats was observed. CT textural features for calcification distribution in the tumors showed that mature cystic teratomas had higher homogeneity and energy than immature teratomas. However, immature teratomas showed higher correlation, entropy, and dissimilarity than mature cystic teratomas among features derived from the gray-level co-occurrence matrix (GLCM) (p < 0.05). No significant differences were observed in the CT features of fats derived from GLCM.

CONCLUSION

Our results demonstrate that calcification distribution on CT is a potential diagnostic biomarker to discriminate mature from immature teratomas, thus enabling optimal therapeutic selection for patients aged < 20 years.

ComBat Harmonization of Myocardial Radiomic Features Sensitive to Cardiac MRI Acquisition Parameters

Purpose

To investigate the effect of ComBat harmonization methods on the robustness of cardiac MRI-derived radiomic features to variations in imaging parameters.

Materials and Methods

This Health Insurance Portability and Accountability Act-compliant retrospective study used a publicly available data set of 11 healthy controls (mean age, 33 years ± 16 [SD]; six men) and five patients (mean age, 52 years ± 16; four men). A single midventricular short-axis section was acquired with 3-T MRI using cine balanced steady-state free precision, T1-weighted, T2-weighted, T1 mapping, and T2 mapping imaging sequences. Each sequence was acquired using baseline parameters and after variations in flip angle, spatial resolution, section thickness, and parallel imaging. Image registration was performed for all sequences at a per-individual level. Manual myocardial contouring was performed, and 1652 radiomic features per sequence were extracted using baseline and variations in imaging parameters. Radiomic feature stability to change in imaging parameters was assessed using Cohen d sensitivity. The stability of radiomic features was assessed both without and after ComBat harmonization of radiomic features. Three ComBat methods were studied: parametric, nonparametric, and Gaussian mixture model (GMM).

Results

For all sequences combined, 51.4% of features were robust to changes in imaging parameters when no ComBat method was applied. ComBat harmonization substantially increased the number of stable features to 95.1% (95% CI: 94.9, 95.3) when parametric ComBat was used and 90.9% (95% CI: 90.6, 91.2) when nonparametric ComBat was used. GMM combat resulted in only 52.6% stable features.

Conclusion

ComBat harmonization improved the stability of radiomic features to changes in imaging parameters across all cardiac MRI sequences.Keywords: Cardiac MRI, Radiomics, ComBat, Harmonization Supplemental material is available for this article. © RSNA, 2023.

Associations between Vertebral Localized Contrast Changes and Adjacent Annular Fissures in Patients with Low Back Pain: A Radiomics Approach

Low back pain (LBP) is multifactorial and associated with various spinal tissue changes, including intervertebral disc fissures, vertebral pathology, and damaged endplates. However, current radiological markers lack specificity and individualized diagnostic capability, and the interactions between the various markers are not fully clear. Radiomics, a data-driven analysis of radiological images, offers a promising approach to improve evaluation and deepen the understanding of spinal changes related to LBP. This study investigated possible associations between vertebral changes and annular fissures using radiomics. A dataset of 61 LBP patients who underwent conventional magnetic resonance imaging followed by discography was analyzed. Radiomics features were extracted from segmented vertebrae and carefully reduced to identify the most relevant features associated with annular fissures. The results revealed three important texture features that display concentrated high-intensity gray levels, extensive regions with elevated gray levels, and localized areas with reduced gray levels within the vertebrae. These features highlight patterns within vertebrae that conventional classification systems cannot reflect on distinguishing between vertebrae adjacent to an intervertebral disc with or without an annular fissure. As such, the present study reveals associations that contribute to the understanding of pathophysiology and may provide improved diagnostics of LBP.

Image harmonization: A review of statistical and deep learning methods for removing batch effects and evaluation metrics for effective harmonization

Magnetic resonance imaging and computed tomography from multiple batches (e.g. sites, scanners, datasets, etc.) are increasingly used alongside complex downstream analyses to obtain new insights into the human brain. However, significant confounding due to batch-related technical variation, called batch effects, is present in this data; direct application of downstream analyses to the data may lead to biased results. Image harmonization methods seek to remove these batch effects and enable increased generalizability and reproducibility of downstream results. In this review, we describe and categorize current approaches in statistical and deep learning harmonization methods. We also describe current evaluation metrics used to assess harmonization methods and provide a standardized framework to evaluate newly-proposed methods for effective harmonization and preservation of biological information. Finally, we provide recommendations to end-users to advocate for more effective use of current methods and to methodologists to direct future efforts and accelerate development of the field.

Neoadjuvant chemotherapy induces an elevation of tumour apparent diffusion coefficient values in patients with ovarian cancer

OBJECTIVES

Multiparametric magnetic resonance imaging (mMRI) is the modality of choice in the imaging of ovarian cancer (OC). We aimed to investigate the feasibility of different types of regions of interest (ROIs) in the measurement of apparent diffusion coefficient (ADC) values of diffusion-weighted imaging in OC patients treated with neoadjuvant chemotherapy (NACT).

METHODS

We retrospectively enrolled 23 consecutive patients with advanced OC who had undergone NACT and mMRI. Seventeen of them had been imaged before and after NACT. Two observers independently measured the ADC values in both ovaries and in the metastatic mass by drawing on a single slice of (1) freehand large ROIs (L-ROIs) covering the solid parts of the whole tumour and (2) three small round ROIs (S-ROIs). The side of the primary ovarian tumour was defined. We evaluated the interobserver reproducibility and statistical significance of the change in tumoural pre- and post-NACT ADC values. Each patient's disease was defined as platinum-sensitive, semi-sensitive, or resistant. The patients were deemed either responders or non-responders.

RESULTS

The interobserver reproducibility of the L-ROI and S-ROI measurements ranged from good to excellent (ICC range: 0.71-0.99). The mean ADC values were significantly higher after NACT in the primary tumour (L-ROI p  < 0.001, S-ROIs p < 0.01), and the increase after NACT was associated with sensitivity to platinum-based chemotherapy. The changes in the ADC values of the omental mass were associated with a response to NACT.

CONCLUSION

The mean ADC values of the primary tumour increased significantly after NACT in the OC patients, and the amount of increase in omental mass was associated with the response to platinum-based NACT. Our study indicates that quantitative analysis of ADC values with a single slice and a whole tumour ROI placement is a reproducible method that has a potential role in the evaluation of NACT response in patients with OC.

TRIAL REGISTRATION

Retrospectively registered (institutional permission code: 5302501; date of the permission: 31.7.2020).

Machine Learning for Brain MRI Data Harmonisation: A Systematic Review

BACKGROUND

Magnetic Resonance Imaging (MRI) data collected from multiple centres can be heterogeneous due to factors such as the scanner used and the site location. To reduce this heterogeneity, the data needs to be harmonised. In recent years, machine learning (ML) has been used to solve different types of problems related to MRI data, showing great promise.

OBJECTIVE

This study explores how well various ML algorithms perform in harmonising MRI data, both implicitly and explicitly, by summarising the findings in relevant peer-reviewed articles. Furthermore, it provides guidelines for the use of current methods and identifies potential future research directions.

METHOD

This review covers articles published through PubMed, Web of Science, and IEEE databases through June 2022. Data from studies were analysed based on the criteria of Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA). Quality assessment questions were derived to assess the quality of the included publications.

RESULTS

a total of 41 articles published between 2015 and 2022 were identified and analysed. In the review, MRI data has been found to be harmonised either in an implicit (n = 21) or an explicit (n = 20) way. Three MRI modalities were identified: structural MRI (n = 28), diffusion MRI (n = 7) and functional MRI (n = 6).

CONCLUSION

Various ML techniques have been employed to harmonise different types of MRI data. There is currently a lack of consistent evaluation methods and metrics used across studies, and it is recommended that the issue be addressed in future studies. Harmonisation of MRI data using ML shows promises in improving performance for ML downstream tasks, while caution should be exercised when using ML-harmonised data for direct interpretation.