Leveraging transformers and explainable AI for Alzheimer’s disease interpretability

Humaira Anzum; Nabil Sadd Sammo; Shamim Akhter

doi:10.1371/journal.pone.0322607

Peer Review History

Original SubmissionSeptember 12, 2024
12 Sep 2024 Author Response https://doi.org/10.1371/journal.pone.0322607.r001
25 Nov 2024 Decision Letter - Laith Alzubaidi, Editor PONE-D-24-39865Leveraging Transformers and Explainable AI for Alzheimer’s Disease InterpretabilityPLOS ONE Dear Dr. Sammo, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Jan 09 2025 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols. We look forward to receiving your revised manuscript. Kind regards, Laith Alzubaidi Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf 2. Please note that PLOS ONE has spec6ific guidelines on code sharing for submissions in which author-generated code underpins the findings in the manuscript. In these cases, all author-generated code must be made available without restrictions upon publication of the work. Please review our guidelines at https://journals.plos.org/plosone/s/materials-and-software-sharing#loc-sharing-code and ensure that your code is shared in a way that follows best practice and facilitates reproducibility and reuse. 3. We note that your Data Availability Statement is currently as follows: All relevant data are within the manuscript and its Supporting Information files. Please confirm at this time whether or not your submission contains all raw data required to replicate the results of your study. Authors must share the “minimal data set” for their submission. PLOS defines the minimal data set to consist of the data required to replicate all study findings reported in the article, as well as related metadata and methods (https://journals.plos.org/plosone/s/data-availability#loc-minimal-data-set-definition). For example, authors should submit the following data: - The values behind the means, standard deviations and other measures reported; - The values used to build graphs; - The points extracted from images for analysis. Authors do not need to submit their entire data set if only a portion of the data was used in the reported study. If your submission does not contain these data, please either upload them as Supporting Information files or deposit them to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. For a list of recommended repositories, please see https://journals.plos.org/plosone/s/recommended-repositories. If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially sensitive information, data are owned by a third-party organization, etc.) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent. If data are owned by a third party, please indicate how others may request data access. 4. Please include your full ethics statement in the ‘Methods’ section of your manuscript file. In your statement, please include the full name of the IRB or ethics committee who approved or waived your study, as well as whether or not you obtained informed written or verbal consent. If consent was waived for your study, please include this information in your statement as well. 5. Please ensure that you refer to Figure 1 in your text as, if accepted, production will need this reference to link the reader to the figure. 6. Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. [Note: HTML markup is below. Please do not edit.] Reviewers' comments: Reviewer's Responses to Questions Comments to the Author 1. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer #1: Partly Reviewer #2: Yes ******** 2. Has the statistical analysis been performed appropriately and rigorously? Reviewer #1: Yes Reviewer #2: Yes ****** 3. Have the authors made all data underlying the findings in their manuscript fully available? The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer #1: Yes Reviewer #2: Yes ****** 4. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer #1: No Reviewer #2: Yes ****** 5. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer #1: The authors of this manuscript present an interesting approach for the interpretability of Alzheimer’s Disease. While the methodology and concept are effective, I have a few suggestions on various sections of the paper that may require improvements to meet the journal's standards. 1. First, I recommend that the authors consider structuring the introduction into several segments based on relevance. While the introduction begins well, the connections between sentences become weaker as it progresses. Additionally, there are insufficient citations for various claims presented in this section. The overall writing quality of the introduction could be improved. For example, the statement, 'However, there is a gap in incorporating XAI techniques into transformer-based models,' is made but is not followed by an explanation or discussion of the gaps, nor is there any citation provided. 2. Similar improvements are needed in the literature review. While the content is generally acceptable, the writing style should be enhanced. It would be beneficial to ensure relevance between sentences and to review the word choices throughout. For instance, if you spell out 'Convolutional Neural Network' once, you can simply use the abbreviation 'CNN' in subsequent references. 3. The results section requires additional validation considering the claims made by the authors. Currently, the authors only performed a comparative analysis with DeepLabV2. I suggest including a few more off-the-shelf network architectures to better validate your results. 4. Quality of Figure 6 and 7 can be improved. 5. Check all the section heading for inaccurate writing format. Reviewer #2: Review of Paper PONE-D-24-39865 Title: Leveraging Transformers and Explainable AI for Alzheimer’s Disease Interpretability Summary: This study addresses Alzheimer’s disease (AD), a progressive brain condition marked by memory loss, cognitive decline, and behavioral changes, affecting one in nine adults over 65. While computer vision applications in MRI-based AD detection have shown accuracies between 80-90%, this study applies advanced transformer models and computer vision to achieve a 99% accuracy rate in categorizing AD stages. By incorporating RNA data and MRI images in near-real-time, this approach improves diagnostic accuracy. The Local Interpretable Model-agnostic Explanations (LIME) method enhances the model's reliability, highlighting essential RNA features and MRI regions for diagnosing AD, thereby making the model’s decisions more interpretable and acceptable for human use. Strengths And Weaknesses: Strengths ● It has a thorough explanation of the main idea. ● Clearly explain the methodology. Weaknesses Overall, the writing can be significantly improved to address the following concerns. ● Introduction: You can add more references in the related work to show that you check new papers. ● Results: The results look good. ****** 6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy. Reviewer #1: No Reviewer #2: No ******** [NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step. Attachments Attachment Submitted filename: Reviewer Comments.pdf Attachment Submitted filename: PONE-D-24-39865_reviewer-comments.docx https://doi.org/10.1371/journal.pone.0322607.r002
Revision 1
31 Jan 2025 Author Response Reviewer 1 Comments and Our Responses 1. Structuring and Writing Quality of the Introduction o Comment: The introduction could be better structured into segments based on relevance. Connections between sentences weaken as it progresses, and insufficient citations are provided. Additionally, the statement regarding the gap in incorporating XAI techniques into transformer-based models lacks explanation or citations. o Response: We have restructured the introduction into clearly defined segments for better readability and logical flow. We have also enhanced the writing quality by refining sentence transitions and word choices. For the statement about the gap in incorporating XAI techniques into transformer-based models, we have provided a brief explanation. We also made an effort to provide the relevant citations. 2. Improvements in the Literature Review o Comment: Similar improvements are needed in the literature review. While the content is generally acceptable, the writing style should be enhanced. It would be beneficial to ensure relevance between sentences and to review the word choices throughout. For instance, if you spell out 'Convolutional Neural Network' once, you can simply use the abbreviation 'CNN' in subsequent references.. o Response: We have carefully reviewed and refined the writing style in the literature review to improve readability and maintain relevance between sentences. Word choices have been optimized, and abbreviations like "Convolutional Neural Network" (CNN) are now consistently used after being spelled out once. 3. Validation of Results Section o Comment: The results section needs further validation beyond the comparative analysis with MobileNetV2. Additional off-the-shelf network architectures should be included for better validation. o Response: We have expanded the results section by including comparative analyses with several additional state-of-the-art architectures, such as Swin Transformers and Skip-connected Convolutional Autoencoders in addition to previously provided Vision Transformers and MobileNetV2. This enhanced validation provides stronger evidence supporting our claims and improves the robustness of our findings. 4. Quality of Figures 6 and 7 o Comment: The quality of Figures 6 and 7 can be improved. o Response: We have improved the quality of Figures 6 and 7 by resizing them and ensuring visual clarity. Labels, legends, and other graphical elements have been adjusted for better comprehensibility. 5. Inaccurate Writing Format in Section Headings o Comment: Check all section headings for formatting inaccuracies. o Response: We have thoroughly reviewed and corrected all section headings to adhere to the journal's formatting guidelines. Inconsistencies in capitalization, numbering, and alignment have been addressed. Attachments Attachment Submitted filename: Response to reviewers.pdf https://doi.org/10.1371/journal.pone.0322607.r003
11 Mar 2025 Decision Letter - Fatih Uysal, Editor PONE-D-24-39865R1Leveraging Transformers and Explainable AI for Alzheimer’s Disease InterpretabilityPLOS ONE Dear Dr. Sammo, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please revise the paper, taking into account both the referee comments and the editor's comments. Please submit your revised manuscript by Apr 25 2025 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols. We look forward to receiving your revised manuscript. Kind regards, Fatih Uysal, Ph.D. Academic Editor PLOS ONE Additional Editor Comments: Please revise the paper, taking into account both the referee comments and the comments I have provided below. 1. There are serious deficiencies in the evaluation metrics required for the evaluation of the classification results in the study. Please obtain the missing Receiver Operating Characteristic (ROC) Curve, the area under the ROC curve (AUC) scores, Cohen's Kappa and Matthews correlation coefficient (MCC) scores completely. 2. It is recommended to make an analysis with several state-of-the-art models in order to compare the results in more depth and to make the proposed approach more prominent. 3. The literature review table needs to be detailed. Here, more new studies related to the current literature should be included and new columns such as "data preprocessing/augmentation, originality, plus and minus aspects" should be added. After this, the difference of this study from the literature and its main contribution to the literature should be expressed in more depth by relating it to the studies in the literature. 4. Detail the information regarding the dataset along with the justifications in terms of traning, validation, test distribution percentages and amounts, data precoessing/augmentation. [Note: HTML markup is below. Please do not edit.] Reviewers' comments: Reviewer's Responses to Questions Comments to the Author 1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer #3: All comments have been addressed Reviewer #4: (No Response) ******** 2. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer #3: Yes Reviewer #4: Yes ****** 3. Has the statistical analysis been performed appropriately and rigorously? Reviewer #3: Yes Reviewer #4: Yes ****** 4. Have the authors made all data underlying the findings in their manuscript fully available? The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer #3: Yes Reviewer #4: Yes ****** 5. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer #3: Yes Reviewer #4: Yes ****** 6. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer #3: The manuscript presents a valuable contribution to Alzheimer’s disease detection using advanced transformers and computer vision techniques. The study is well-structured and demonstrates impressive accuracy. To further enhance the work, we recommend citing "Detection of Alzheimer’s and Parkinson’s Diseases Using Deep Learning-Based Various Transformers Models" (https://doi.org/10.3390/engproc2024073004) to provide additional context and strengthen the research foundation. Overall, we suggest minor revisions and look forward to the updated submission. Reviewer #4: Your manuscript offers a valuable contribution to AI-driven Alzheimer’s disease diagnosis, demonstrating high accuracy and integrating explainability via LIME. The combination of transformers for MRI and RNA sequencing analysis is innovative and well-justified. The literature review is comprehensive, and the methodological framework is clearly presented. However, the manuscript could benefit from additional validation on external datasets, a more detailed discussion on clinical applicability, and minor refinements in writing and formatting. These improvements would enhance the paper’s clarity and real-world impact :- 1. Introduction & Motivation The introduction clearly states the importance of early AD diagnosis and the role of AI in medical imaging, but the research gap could be more explicitly defined to emphasize why transformers are superior to CNNs or traditional machine learning models. Additionally, the claim that "there is no cure for AD" should be softened, as some experimental treatments exist. A brief outline of the paper structure at the end of the introduction would improve readability and guide the reader. 2. Methodology & Model Description The methodology section is highly technical and would benefit from a flowchart or diagram illustrating the proposed model architecture. Additionally, key training parameters (batch size, optimizer, number of epochs) should be summarized in a table to improve clarity. The explanation of LIME should be expanded to include how interpretability was validated—was expert medical review conducted to confirm whether the AI highlights relevant features? 3. Results & Validation The reported accuracy of 99% raises concerns about overfitting, especially since deep learning models typically struggle with medical imaging data generalization. The study should include external validation using an independent dataset (e.g., ADNI, OASIS) or a detailed description to confirm robustness. Additionally, statistical measures such as confidence intervals or standard deviations should be included to ensure the reliability of the reported results. While the comparative analysis with other models is strong, it could be further improved by incorporating benchmarks such as ResNet or EfficientNet. 4. Figures & Tables Figures 6 and 7 require more detailed captions explaining the significance of their findings. The LIME heatmaps should include clinical validation, ideally with references to known AD biomarkers. Additionally, the results section would benefit from an ROC-AUC curve to provide a more comprehensive evaluation of classification performance. 5. Discussion & Interpretability The discussion should better address dataset biases, such as demographic imbalances and variations in MRI scanning equipment. The interpretability aspect of LIME should be assessed from a clinical perspective—does the AI highlight regions relevant to AD diagnosis? Furthermore, potential regulatory challenges (e.g., FDA approval for AI-based diagnostic tools) should be briefly discussed. 6. Writing & Formatting Improvements Some sections contain long, complex sentences that could be restructured for clarity. The use of abbreviations should be consistent (e.g., CNN should be abbreviated after its first mention). Additionally, minor grammatical redundancies, such as "AI-based artificial intelligence," should be corrected for conciseness. ****** 7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy. Reviewer #3: No Reviewer #4: Yes: Aasim Ayaz Wani ******** [NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step. Attachments Attachment Submitted filename: Reviewer Comment.docx https://doi.org/10.1371/journal.pone.0322607.r004
Revision 2
24 Mar 2025 Author Response Responses and Actions to Reviewer Comments Dear Editor and Reviewers, The authors would like to thank the reviewers for providing insightful comments that helped to make the journal paper better. All review judgments and comments are looked at by the authors. Following the reviewers' recommendations, they also offer remarks, feedback, and necessary changes to the work. The authors express their gratitude to all of the esteemed reviewers for their hard work. Best Regards, Authors, Manuscript Number: PONE-D-24-39865 ============== Review Comments and Corresponding Actions ====================== Additional Editor Comments: Comment-1: There are serious deficiencies in the evaluation metrics required for the evaluation of the classification results in the study. Please obtain the missing Receiver Operating Characteristic (ROC) Curve, the area under the ROC curve (AUC) scores, and Cohen's Kappa and Matthews correlation coefficient (MCC) scores completely. Response-1: Thank you very much for your comments. We add the Cohen's Kappa, Matthews correlation coefficient (MCC) scores, and AUC values of the experimental models in Table 5. AUCs of all classes are generated from the Receiver Operating Characteristic (ROC) curves in Figure 4 using a vision transformer. The following lines are included in Section 5.1, line 352: “Table 6 outlines the outcomes obtained by employing a vision transformer across various classes in the testing dataset. The NonDemented class attains the highest accuracy at 99.7%, while the MildDemented class registers the lowest accuracy at 97.31%. The vision transformer accurately classifies 98.6% of cases, with only a few instances of mis-prediction. The accuracy of the MildDemented class is less due to the presence of fewer training images in the dataset. The model performs a little less when there is a small number of training data. Other classes achieve appreciable results. Consequently, the overall accuracy of the model stands at 98.6% for all testing images, with a 95% confidence interval (CI) of [97.41%, 99.36%], indicating a high degree of reliability in classification performance. We derive the CI using 5-fold cross-validation, where the dataset was split into five subsets, training and testing the model iteratively to ensure stable performance across folds. The narrow CI range confirms robustness, as it reflects consistent accuracy with minimal variation across these folds. The CI provides a lower and upper range because it accounts for statistical uncertainty in the sample data, estimating the plausible bounds of the true accuracy. The lower bound (97.41%) represents the minimum expected performance with 95% confidence, while the upper bound (99.36%) suggests the potential peak accuracy under optimal conditions. The corresponding loss and accuracy curves are depicted in Fig. 3 and Fig. 2. The accuracy curves for both training and testing exhibit an initial surge to 80% within the initial 5 epochs, eventually surpassing 95% after 15 epochs. Both curves demonstrate similar trends in loss calculations, reaching saturation after 15 epochs at the same point. Also, the ROC curve is displayed in Fig. 4. The Cohen Kappa coefficient value we obtain is 0.97, indicating almost perfect agreement between the true and predicted labels. Similarly, the Matthews Correlation Coefficient (MCC) is 0.98, reflecting a strong correlation and robust performance of the model across all classes. In Fig. 5, LIME explanations are presented for all four classes of test images. The LIME explanation highlights the specific regions of the images that contributed to the prediction of these classes. Red patches in the explanation images represent areas that positively influenced the model’s accurate prediction. Positive influence implies that these regions were instrumental in the model correctly identifying the class or label. In essence, the presence of red zones indicates that the characteristics or patterns in those areas align with the expected class. In Fig. 6, overlays of the explanations are displayed. The yellow lines point out distinct pixels that played significant roles in predicting the class of these particular images.” Comment-2: It is recommended to analyze several state-of-the-art models to compare the results in more depth and to make the proposed approach more prominent. Response-2: Two new state-of-the-art models, including ResNet 50, and EfficientNet Models, are experimented with, and their results are presented in Table 5. The following lines are included in Section 5.2, line 406: “Additionally, EfficientNet-B0 and ResNet-50 are included for benchmarking, achieving accuracies of 91.57% and 86.59%, respectively, further validating the robustness of our approach. These results collectively highlight the trade-offs between accuracy, computational efficiency, and deployment feasibility. Based on our findings, it can be conclusively stated that the ViT proves to be a highly effective and efficient algorithm for the classification of MRI images related to Alzheimer’s Disease. Table 5 displays the Accuracy, F1-score, Cohen’s Kappa, and Matthew’s correlation coefficient of each model on the dataset. Based on our findings, it can be conclusively stated that the Vision Transformer proves to be a highly effective and efficient algorithm for the classification of MRI images related to Alzheimer’s Disease. Also, Fig. 4 represents the ROC curve of ViT where we can see that ViT achieves high Area Under the Curve (AUC) scores across all Alzheimer’s Disease classes, reflecting its strong ability to distinguish between positive and negative cases for each category. The AUC values 0.98 for ModerateDemented, 0.96 for MildDemented, 1.00 for NonDemented, and 0.99 for VeryMildDemented indicate near-perfect discrimination, with NonDemented reaching an ideal 1.00, aligning with its top accuracy of 99.70%. These results underscore ViT’s robustness in handling class-specific patterns in brain imaging data, even for challenging cases like MildDemented as the number of training images are very low, where accuracy dips slightly and so does the AUC. Also, the Cohen’s Kappa and Matthew’s correlation coefficients are consistent with these findings.” Comment-3: The literature review table needs to be detailed. Here, more new studies related to the current literature should be included, and new columns such as "data preprocessing/augmentation, originality, plus and minus aspects" should be added. After this, the difference of this study from the literature and its main contribution to the literature should be expressed in more depth by relating it to the studies in the literature. Response-3: Three new related studies are added accordingly, and the literature review table is split into two separate tables to cover all these aspects mentioned (originality, plus and minus aspects). In the Reference section, the following three references are added: 30. Guven M. Detection of Alzheimer’s and Parkinson’s Diseases Using Deep Learning-Based Various Transformers Models. Eng. Proc. 2024;73(1):4. Presented at the 4th International Electronic Conference on Biosensors, 20–22 May 2024. 629 Available online: https://sciforum.net/event/IECB2024. https://doi.org/10.3390/engproc2024073004. 31. Liu M, Li F, Yan H, Wang K, Ma Y, Shen L, A deep learning system for differential diagnosis of Alzheimer’s disease and mild cognitive impairment using structural MRI. NeuroImage. 2021;225:117496. https://doi.org/10.1016/j.neuroimage.2020.117496. 32. Wang H, Shen Y, Wang S, Xiao T, Deng L, Wang X, Multimodal deep learning for Alzheimer’s disease dementia assessment. Nature Communications. 2020;11(1):1–9. https://doi.org/10.1038/s41467-020-18187-0. In addition, the following lines are added in Section 2 Literature Review line 116: “Wang et al. [32] developed a multimodal deep learning framework for Alzheimer’s Disease dementia assessment, integrating data from neuroimaging, genetic markers, and cognitive tests. Their approach combines CNNs and recurrent neural networks (RNNs) to capture both spatial and temporal patterns in the data. The study shows that multimodal data fusion significantly improves diagnostic accuracy compared to single-modality approaches. The authors reported an accuracy of 92.1% for Alzheimer’s Disease (AD) diagnosis using their multimodal deep learning framework, which integrated neuroimaging, genetic markers, and cognitive tests. This multi-modal approach was quite a new addition to the field of AD diagnosis at the time of publication, but new state-of-the-art algorithms like models can be used for further accuracy improvements. Another study conducted by Liu M et al. [31] proposed a deep learning system for the differential diagnosis of Alzheimer’s Disease (AD) and mild cognitive impairment (MCI) using structural MRI. Their model leverages 3D convolutional neural networks (CNNs) to extract features from brain scans, achieving high accuracy in distinguishing between AD, MCI, and healthy controls. The study highlights the potential of deep learning in automating AD diagnosis and emphasizes the importance of structural MRI as a key biomarker for early detection. reported an accuracy of 88.6% for differentiating Alzheimer’s Disease (AD) from healthy controls and 76.5% for distinguishing mild cognitive impairment (MCI) from healthy controls using their deep learning system based on structural MRI. The accuracy remains relatively low due to the use of CNN-based architectures, and the usability of the moderately accurate model can cause a serious discrepancy in the field of AD detection.” The following lines are added in Section 2 Literature Review line 151: “Recent work by [30] applied transformer-based models, including Swin Transformer, Vision Transformer (ViT), and Bidirectional Encoder Representation from Image Transformers (BEiT), to classify Alzheimer’s and Parkinson’s diseases using brain imaging data. The study utilized a balanced dataset of 450 brain images, achieving classification accuracy exceeding 80%, with ViT demonstrating the highest performance (94.4% accuracy, 94.7% precision). While the results highlight the efficacy of transformer architectures in disease detection, the study has notable shortcomings. The dataset size (450 images) is relatively small, which may limit the generalizability of the findings. Also, the accuracy is not satisfactory in terms of AD disease diagnosis.” Table 1 is partitioned into two parts, Table 1 and Table 2, to add suggested new columns such as "data preprocessing/augmentation, originality, plus and minus aspects." Comment-4: Detail the information regarding the dataset along with the justifications in terms of training, validation, test distribution percentages and amounts, and data preprocessing/augmentation. Response-4: We add detail information on the dataset along with the justifications in terms of training, validation, test distribution percentages, and amounts. The following part is added to the paper in the 3.1 section, page 7: “We utilized the Alzheimer’s dataset from Kaggle [21], which includes gray MRI scans of the brain from individuals in various stages of Alzheimer’s disease. The dataset encompasses four classes: MildDemented, ModerateDemented, NonDemented, and VeryMildDemented. It is organized into two folders: Train and Test, comprising a total of 5121 training images across the four classes and 1379 test images. Table 3 illustrates the distribution of images in the dataset. This dataset is valuable for predicting Alzheimer’s disease stages using computer vision algorithms. Given the limited size of the test set, we allocated 10% of the training dataset for validation to ensure a robust evaluation of the model’s performance. This approach was necessary because the dataset did not provide a separate validation set, and splitting the training set further would have risked overfitting due to the imbalanced class distribution. Using a small portion of the test set for validation allowed us to tune hyper-parameters and monitor model performance during training without significantly compromising the final evaluation on the remaining test data.” Reviewer#3 Comments: Comment-1: The manuscript presents a valuable contribution to Alzheimer’s disease detection using advanced transformers and computer vision techniques. The study is well structured and demonstrates impressive accuracy. To further enhance the work, we recommend citing "Detection of Alzheimer’s and Parkinson’s Diseases Using Deep Learning-Based Various Transformers Models" (https://doi.org/10.3390/engproc2024073004) to provide additional context and strengthen the research foundation. Overall, we suggest minor revisions and look forward to the updated submission. Response-1: We appreciate your constructive suggestion. Thank you. The paper titled "Detection of Alzheimer’s and Parkinson’s Diseases Using Deep Learning-Based Various Transformers Models" (https://doi.org/10.3390/engproc2024073004) has been studied rigorously to establish context and strengthen the foundation of this study. Also, it has been added to the reference section. In the paper page 20 in the References section, we add the following references: 30. Guven M. Detection of Alzheimer’s and Parkinson’s Diseases Using Deep Learning-Based Various Transformers Models. Eng. Proc. 2024;73(1):4. Presented at the 4th International Electronic Conference on Biosensors, 20–22 May 2024. Available online: https://sciforum.net/event/IECB2024. https://doi.org/10.3390/engproc2024073004. In the paper in the Literature Review Section, page 4, line 151, we add the following paragraph: “Recent work by [30] applied transformer-based models, including Swin Transformer, Vision Transformer (ViT), and Bidirectional Encoder Representation from Image Transformers (BEiT), to classify Alzheimer’s and Parkinson’s diseases using brain imaging data. The study utilized a balanced dataset of 450 brain images, achieving classification accuracy exceeding 80%, with ViT demonstrating the highest performance (94.4% accuracy, 94.7% precision). While the results highlight the efficacy of transformer architectures in disease detection, the study has notable shortcomings. The dataset size (450 images) is relatively small, which may limit the generalizability of the findings. Also, the accuracy is not satisfactory in terms of AD disease diagnosis.” Reviewer#4 Comments: Your manuscript offers a valuable contribution to AI-driven Alzheimer’s disease diagnosis, demonstrating high accuracy and integrating explainability via LIME. The combination of transformers for MRI and RNA sequencing analysis is innovative and well-justified. The literature review is comprehensive, and the methodological framework is presented. However, the manuscript could benefit from additional validation on external datasets, a more detailed discussion on clinical applicability, and minor refinements in writing and formatting. These improvements would enhance the paper’s clarity and real-world impact: Comment-1: The introduction clearly states the importance of early AD diagnosis and the role of AI in medical imaging, but the research gap could be more explicitly defined to emphasize why transformers are superior to CNNs or traditional machine learning models. Additionally, the claim that "there is no cure for AD" should be softened, as some experimental treatments exist. A brief outline of the paper structure at the end of the introduction would improve readability and guide the reader. Response-1: Thank you for your valuable suggestions. We have revised the introduction to explicitly define the research gap, emphasizing the advantages of transformers over CNNs and traditional machine learning models. The statement "there is no cure for AD" is replaced as “Currently there is almost no cure for AD except very few experimental treatments.” The following paragraph is added in the Introduction section, page 2, 32 lines: “A notable advancement in DL is the development of the Transformer architecture, which employs attention mechanisms to capture relationships between elements in data. Initially designed for Natural Language Processing (NLP), the Transformer’s architecture has been adapted for image analysis, Attachments Attachment Submitted filename: Response to Reviewers.docx https://doi.org/10.1371/journal.pone.0322607.r005
26 Mar 2025 Decision Letter - Fatih Uysal, Editor Leveraging Transformers and Explainable AI for Alzheimer’s Disease Interpretability PONE-D-24-39865R2 Dear Dr. Sammo, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice will be generated when your article is formally accepted. Please note, if your institution has a publishing partnership with PLOS and your article meets the relevant criteria, all or part of your publication costs will be covered. Please make sure your user information is up-to-date by logging into Editorial Manager at Editorial Manager® and clicking the ‘Update My Information' link at the top of the page. If you have any questions relating to publication charges, please contact our Author Billing department directly at authorbilling@plos.org. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org. Kind regards, Fatih Uysal, Ph.D. Academic Editor PLOS ONE Additional Editor Comments (optional): After considering the reviewer comments and evaluating the quality of the paper, it has been decided to accept it due to its potential to contribute to the literature and its final form. Reviewers' comments: https://doi.org/10.1371/journal.pone.0322607.r006
Formally Accepted
Acceptance Letter - Fatih Uysal, Editor PONE-D-24-39865R2 PLOS ONE Dear Dr. Sammo, I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now being handed over to our production team. At this stage, our production department will prepare your paper for publication. This includes ensuring the following: * All references, tables, and figures are properly cited * All relevant supporting information is included in the manuscript submission, * There are no issues that prevent the paper from being properly typeset If revisions are needed, the production department will contact you directly to resolve them. If no revisions are needed, you will receive an email when the publication date has been set. At this time, we do not offer pre-publication proofs to authors during production of the accepted work. Please keep in mind that we are working through a large volume of accepted articles, so please give us a few weeks to review your paper and let you know the next and final steps. Lastly, if your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org. If we can help with anything else, please email us at customercare@plos.org. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Fatih Uysal Academic Editor PLOS ONE https://doi.org/10.1371/journal.pone.0322607.r007

Open letter on the publication of peer review reports

PLOS recognizes the benefits of transparency in the peer review process. Therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. Reviewers remain anonymous, unless they choose to reveal their names.

We encourage other journals to join us in this initiative. We hope that our action inspires the community, including researchers, research funders, and research institutions, to recognize the benefits of published peer review reports for all parts of the research system.

Learn more at ASAPbio .