Training Artificial Neural Networks to Detect Multiple Sclerosis Lesions Using Granulometric Data from Preprocessed Magnetic Resonance Images with Morphological Transformations
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe symptoms of multiple sclerosis (MS) are determined by demyelinating lesions in the brain and spinal cord, typically diagnosed using magnetic resonance imaging (MRI). However, MRI can include undesirable components like noise or intensity variations. This paper proposes an MRI preprocessing algorithm using mathematical morphology (MM) to filter out these components and extract relevant structures, including MS lesions. The algorithm computes granulometry to describe lesion sizes and trains artificial neural networks (ANNs) to predict MS diagnosis, achieving high accuracy (0.9753) and low cross-entropy loss (0.0247), thus aiding specialists in diagnosing and assessing MS progression. However, I still have some concerns.
1. The motivation for lesion detection is not clear. The authors should provide a more detailed discussion on this topic.
2. The introduction to segmentation is not clear. The authors should cover several relevant topics, including fully-supervised, semi-supervised, and unsupervised medical image segmentation.
3. The authors seem to have missed some relevant literature. Specifically, they don't discuss learning-based methods for segmentation tasks, missing out on several relevant citations, e.g. “Class-aware adversarial transformers for medical image segmentation”, "Momentum contrastive voxel-wise representation learning for semi-supervised volumetric medical image segmentation", "SimCVD: Simple contrastive voxel-wise representation distillation for semi-supervised medical image segmentation", "Incremental learning meets transfer learning: Application to multi-site prostate MRI segmentation", “Bootstrapping semi-supervised medical image segmentation with anatomical-aware contrastive distillation”, "Unsupervised wasserstein distance guided domain adaptation for 3d multi-domain liver segmentation", “Mine your own anatomy: Revisiting medical image segmentation with extremely limited labels”, “Rethinking Semi-Supervised Medical Image Segmentation: A Variance-Reduction Perspective”, “ACTION++: Improving Semi-supervised Medical Image Segmentation with Adaptive Anatomical Contrast”, and “Implicit Anatomical Rendering for Medical Image Segmentation with Stochastic Experts”. These methods are relevant to the method proposed in this paper. These relevant papers should be included in the reference list.
Comments on the Quality of English LanguageThe authors should improve the grammar and writing style.
Author Response
"Please see the attachment."
Author Response File: Author Response.docx
Reviewer 2 Report
Comments and Suggestions for AuthorsThis manuscript proposes an MRI preprocessing algorithm using mathematical morphology and granulometry to enhance the detection and characterization of MS lesions. The authors introduce a method combining granulometry with an ANN model, which improves the accuracy of MS lesion detection and size estimation. The manuscript is well-structured, with clear explanations of methods and results. However, there are still areas that need improvement.
Major Concerns:
1. Please supplement the description of the evaluation metrics in Figures 11-13 and Table 6, including the confusion matrix and cross-entropy loss function. What do they represent, and what conclusions can be drawn from the figure and table?
2. What conclusions can be drawn from Figures 16 and 17? Please provide additional explanations.
3. Table 7 compares the performance results of several studies in analyzing multiple sclerosis lesions in MRI. What conclusions can be drawn from these comparisons, and how does the article demonstrate the superiority of its proposed method? Please compare using the same evaluation metrics.
Minor Concerns:
1. In Section 2.6, Tables 1 and 2 already introduce the algorithm process. The content from lines 207 to 246 in the text is redundant.
2. Section 2.6 introduces the new algorithm proposed in the article. It is recommended to move this section to the beginning of Section 2 to allow readers to understand the main work of the article more quickly.
3. Reference 30 is missing the issue number.
4. Most of the cited references are relatively old. It is recommended to strengthen the research on the latest literature from the past three years.
5.There are too many figures and tables, making the content appear redundan. For example, Tables 4, 5, and Figure 10 describe the same data.
6. Lines 343-348: Please supplement the relevant references.
Author Response
"Please see the attachment."
Author Response File: Author Response.docx