Abstract:In this work, we present a comparison of a shallow and a deep learning architecture for the automated segmentation of white matter lesions in MR images of multiple sclerosis patients. In particular, we train and test both methods on early stage disease patients, to verify their performance in challenging conditions, more similar to a clinical setting than what is typically provided in multiple sclerosis segmentation challenges. Furthermore, we evaluate a prototype naive combination of the two methods, which refines the final segmentation. All methods were trained on 32 patients, and the evaluation was performed on a pure test set of 73 cases. Results show low lesion-wise false positives (30%) for the deep learning architecture, whereas the shallow architecture yields the best Dice coefficient (63%) and volume difference (19%). Combining both shallow and deep architectures further improves the lesion-wise metrics (69% and 26% lesion-wise true and false positive rate, respectively).

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: How do the techniques for automatic segmentation of white matter lesions perform in the early stage of multiple sclerosis (MS)? Specifically, the author compared the performance of shallow - learning architectures and deep - learning architectures in this task and verified their performance under challenging clinical conditions (such as small lesions and low lesion loads). ### Main problems: 1. **Segmentation of small lesions in early - stage MS patients**: Most of the existing evaluations are carried out in patients with high lesion loads, while the lesions in early - stage MS patients are usually small and few in number, which poses higher requirements for automated segmentation methods. 2. **Comparison between shallow - and deep - learning architectures**: By comparing the performance of shallow - learning (such as k - NN combined with partial volume modeling) and deep - learning (such as 3D convolutional neural networks, CNNs) in early - stage MS patients, explore which method is more suitable for handling these challenging cases. 3. **Effect of method combination**: Explore whether combining shallow - and deep - learning methods can further improve the segmentation performance, especially in terms of reducing the false - positive rate and increasing the true - positive rate. ### Research background: - **Multiple sclerosis (MS)** is a demyelinating disease that affects the central nervous system, resulting in focal lesions in the white matter. - **Magnetic resonance imaging (MRI)** is an important tool for diagnosing and monitoring the progress of MS and the response to treatment. - **Manual annotation** is considered the clinical gold standard for MS lesion identification, but it is time - consuming and susceptible to inter - observer variability. - **Automated methods**, especially supervised learning techniques, have performed well in MS lesion detection, and deep - learning architectures have also made significant progress in recent years. ### Main contributions of the paper: - **Dataset selection**: Use a dataset of early - stage MS patients to be closer to the actual clinical situation. - **Method comparison**: Compare in detail the performance of shallow - learning and deep - learning architectures under different minimum lesion sizes and total lesion loads. - **Method combination**: Propose a simple method combination (PV - CNNs) and verify its effect through experiments. ### Conclusions: - **Shallow - learning (LeMan - PV)** performs best in terms of Dice coefficient and volume difference, but is not as good as deep - learning in reducing the false - positive rate. - **Deep - learning (CNNs)** performs well in reducing the false - positive rate, but its overall segmentation performance is slightly inferior to that of shallow - learning. - **Combination method (PV - CNNs)** is superior to single methods in terms of the true - positive rate and false - positive rate at the lesion level, but performs poorly in terms of volume difference. Through these studies, the author provides valuable insights for the automatic segmentation of white matter lesions in early - stage MS patients and points out the directions for future improvement.

Shallow vs deep learning architectures for white matter lesion segmentation in the early stages of multiple sclerosis

Contrast-Enhancing Lesion Segmentation in Multiple Sclerosis: A Deep Learning Approach Validated in a Multicentric Cohort

Consensus of algorithms for lesion segmentation in brain MRI studies of multiple sclerosis

Multiple Sclerosis Lesion Segmentation from Brain MRI via Fully Convolutional Neural Networks

Automated Segmentation of Multiple Sclerosis Lesions using Deep Learning

Deep Learning on Conventional Magnetic Resonance Imaging Improves the Diagnosis of Multiple Sclerosis Mimics

LST-AI: A deep learning ensemble for accurate MS lesion segmentation

Segmentation of Cerebral Small Vessel Diseases-White Matter Hyperintensities Based on a Deep Learning System

Minimizing the effect of white matter lesions on deep learning based tissue segmentation for brain volumetry

Multiple Sclerosis Lesion Analysis in Brain Magnetic Resonance Images: Techniques and Clinical Applications

Investigation of an efficient multi-modal convolutional neural network for multiple sclerosis lesion detection

Acute and sub-acute stroke lesion segmentation from multimodal MRI

Compound attention embedded dual channel encoder-decoder for ms lesion segmentation from brain MRI

Image Segmentation with Priority Based Apposite Feature Extraction Model for Detection of Multiple Sclerosis in MR Images Using Deep Learning Technique

Toward Generalizable Multiple Sclerosis Lesion Segmentation Models

Machine learning to investigate superficial white matter integrity in early multiple sclerosis

Enhanced image quality and lesion detection in FLAIR MRI of white matter hyperintensity through deep learning-based reconstruction

Interpretable deep learning as a means for decrypting disease signature in multiple sclerosis

Axial multi-layer perceptron architecture for automatic segmentation of choroid plexus in multiple sclerosis

Longitudinal Segmentation of MS Lesions via Temporal Difference Weighting