List of Papers Browse by Subject Areas Author List
Abstract
Transcranial Direct Current Stimulation (tDCS) is a non-invasive brain stimulation method that applies neuromodulatory effects to the brain via low-intensity, direct current. It has shown possible positive effects in areas such as depression, substance use disorder, anxiety, and pain. Unfortunately, mixed trial results have delayed the field’s progress. Electrical current field approximation provides a way for tDCS researchers to estimate how an individual will respond to specific tDCS parameters. Publicly available physics-based stimulators have led to much progress; however, they can be error-prone, susceptible to quality issues (e.g., poor segmentation), and take multiple hours to run. Digital functional twins provide a method of estimating brain function in response to stimuli using computational methods. We seek to implement this idea for individualized tDCS. Hence, this work provides a proof-of-concept for generating electrical field maps for tDCS directly from T1-weighted magnetic resonance images (MRIs). Our deep learning method employs special loss regularizations to improve the model’s generalizability and
calibration across individual scans and electrode montages. Users may enter a desired electrode montage in addition to the unique MRI for a custom output. Our dataset includes 442 unique individual heads from individuals across the adult lifespan. The pipeline can generate results on the scale of minutes, unlike physics-based systems that can take 1-3 hours. Overall, our methods will help streamline the process of individual current dose estimations for improved tDCS interventions.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/1533_paper.pdf
SharedIt Link: https://rdcu.be/dV1PQ
SpringerLink (DOI): https://doi.org/10.1007/978-3-031-72069-7_44
Supplementary Material: N/A
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{Sto_Towards_MICCAI2024,
author = { Stolte, Skylar E. and Indahlastari, Aprinda and Albizu, Alejandro and Woods, Adam J. and Fang, Ruogu},
title = { { Towards tDCS Digital Twins using Deep Learning-based Direct Estimation of Personalized Electrical Field Maps from T1-Weighted MRI } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15002},
month = {October},
page = {465 -- 475}
}
Reviews
Review #1
- Please describe the contribution of the paper
The authors highlight the need for personalized electrical field map estimation for tDCS. According to the authors a potential limitation of some tDCS intervention trials is that they did not account for individual anatomy. Personalized electrical field map estimation is currently accomplished via finite element analysis, but this is time consuming. Therefore, the authors propose using deep learning to estimate the field maps quickly.
- Please list the main strengths of the paper; you should write about a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
If timely estimation of personalized electrical field maps is for tDCS research and use, then the solution proposed by the authors appears to be reasonable. The authors used a larger dataset and a different methodology than had previously been used by a different deep learning solution attempting to solve the same problem. The authors optimized their model training by implementing multiple loss functions.
- Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
The model architecture and training methodology is not novel. The validation of the model performance is limited to an ablation study and qualitative comparison of predictions. It is not clear if the model performs well enough for application purposes.
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Do you have any additional comments regarding the paper’s reproducibility?
N/A
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
The text in Figure 1 is too small to read when printed. Page 5: “T1 MRI” should be “T1-weighted MRI”.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making
Weak Reject — could be rejected, dependent on rebuttal (3)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper lacks novelty, but appears to have clinical utility. However, it is not clear if the methodology performs well enough to achieve clinical utility.
- Reviewer confidence
Somewhat confident (2)
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
N/A
- [Post rebuttal] Please justify your decision
N/A
Review #2
- Please describe the contribution of the paper
The authors propose a novel deep learning methodology aimed at generating electric field maps using high-resolution T1 images obtained from healthy volunteers. This advancement presents a significant improvement over existing physics-based methods, which typically rely on time-consuming simulations performed by tools such as ROAST or SimNibs, with simulation times ranging from 1 to 3 hours. By employing a dual-learning approach integrating SwinUNETR for regression (to predict electric field maps) and EfficientNet for classification (to generate electric montage simulations), the model demonstrates the capability to extract digital twins from T1-weighted MRI images and estimate changes in electric current flows based on two distinct electrical montages.
- Please list the main strengths of the paper; you should write about a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
The proposed deep learning approach adopts a sophisticated calibration strategy, harnessing both regression and classification outputs to optimize accuracy. An ablation study underscores the efficacy of this intricate validation scheme in enhancing model performance. This innovative methodology enables the simulation of potential changes in electric fields based on digital functional twins, marking a significant departure from traditional approaches. However, it’s crucial to note that while promising, this technique has yet to undergo comprehensive validation in clinical practice. In light of the current state-of-the-art in models addressing similar problems, this work represents a notable advancement, primarily attributed to its ability to drastically reduce the time required for modeling electric current field maps. By streamlining this process, the model not only enhances efficiency but also opens avenues for more widespread adoption and exploration of transcranial Direct Current Stimulation (tDCS) in various research and clinical applications. For broader acceptance and applicability, thorough validation of the method across different patient cohorts and imaging machines is imperative. This validation process will not only bolster confidence in the model’s performance but also ensure its robustness and reliability across diverse real-world scenarios. Additionally, validation across various imaging platforms and populations will help identify potential biases and ensure the model’s generalizability and effectiveness in clinical settings.
- Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
The acknowledgment of transcranial direct stimulation (tDCS) as a promising technique lacking conclusive evidence of efficacy within evidence-based medicine sets the context for evaluating the presented work. While the study represents a notable progression in tDCS research, its potential impact on demonstrating the technique’s efficacy across various diseases remains speculative. Comparing the proposed method with the established gold standard, ROAST—a physics-based approach for estimating electric field maps—adds credibility to the research. ROAST’s validation, involving manual brain segmentation, underscores its reliability and robustness, as demonstrated in the manuscript detailing its development and comparison with alternative methods. However, assessing the clinical significance of differences in electric field maps estimation based solely on reported metrics poses a challenge. Metrics like Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Structural Similarity Index (SSIM), and Expected Calibration Error (ECE) offer quantitative insights into model performance. Yet, translating these metrics into clinical relevance requires further investigation. Moreover, the method exhibits poorer performance in young adults, akin to the previous ROAST method. While the authors note that ROAST encountered difficulties in a percentage of young patients, it remains unclear to what extent this method might be unsuitable for this demographic. Clarifying the degree of applicability to this population would provide valuable insights into the method’s limitations and potential areas for improvement.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not provide sufficient information for reproducibility.
- Do you have any additional comments regarding the paper’s reproducibility?
In some instances, understanding what is being measured and how it is measured proves challenging. For example, the authors should clarify what is being assessed in “MAE” since the reviewer found it difficult to discern the exact metric under consideration. Additionally, it is crucial to establish to what extent the reported metrics reflect clinically significant differences. Specifically, elucidating how inaccuracies in the methods impact the metrics used for tDCS planning and stimulation would provide valuable context. Without this clarification, forming an opinion on whether the current methods offer an advantage over previously published works becomes challenging, beyond the apparent benefit of time-saving—an aspect that might not be a significant obstacle given the non-emergent nature of the diseases under study.
The current work is tested in two different cohorts: one adult cohort from a clinical trial and one young cohort from the Human Connectome Project’s (HCP) Young Adult study. The code is not provided.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
I congratulate the authors for their commendable work and the significant improvement they’ve achieved over existing methods for the addressed problem. However, to enhance clarity for readers not well-versed in the field, I would suggest incorporating some clarifications and adjustments to the conclusions section: -Provide the percentage of patients in which ROAST failed, and report whether the authors attempted their approach in the demographic where ROAST encountered difficulties. This information would shed light on the method’s adaptability across different patient populations. -Offer insights into what is being measured and the nature of the error quantified in the reported metrics. Providing at least some level of detail about these aspects would help readers better understand how to interpret the reported accuracy and assess any potential biases introduced by inaccuracies in the proposed method. -Regarding the conclusions section, while acknowledging the advantages of the proposed method is important, t it might appear repetitive to simply reiterate praise already mentioned in the manuscript. Instead, I suggest using this section to provide a more objective and critical assessment of the work. This could involve discussing limitations, potential areas for improvement, and implications for future research, thus offering a more balanced perspective on the findings.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making
Weak Accept — could be accepted, dependent on rebuttal (4)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The authors provide a reliable method for electric field maps estimation based T1-weighted MRIs in healthy subjects. The method offers advantages over previous methods (ie ROAST), mainly in time. The author could provide a more critical view of their findings and acknowledge that currently, tDCS has not proven to be effective in any field, so the application of their work in clinical practice is hypothetical
- Reviewer confidence
Somewhat confident (2)
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
N/A
- [Post rebuttal] Please justify your decision
N/A
Review #3
- Please describe the contribution of the paper
This paper introduces the first deep learning based method that predicts tDCS electric field maps directly from T1 MRI images. It improves upon the previous development [14] by training functional digital twins of fMRIs with a larger training sample size and an improved model structure (SwinUNETR). The contribution is a proof-of-concept for the goal of real-time prediction of tDCS electric field maps in the future. This has clinical significance, because individualized tDCS is an unmet CAI need given the limitation of conventional, fixed-current fixed-electrode-placement tDCS.
- Please list the main strengths of the paper; you should write about a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
The work is an innovative application of SwinUNETR transformer model. The results provide a foundation for future development with improved accuracy and generalizability. The advantage includes the reduction of prediction to within minutes, which is a major improvement from state-of-the-art physics-based toolboxes. Overall, the work has potential impact on future development on individualized current dose and electrode placement for tDCS.
- Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
The study is limited to only two electrode montages, which may be sufficient as a proof-of-concept but leaves room for improvement.
The training data combines young and old age groups, whereas the test performance is better on older adult data. The trained model also demonstrates different error patterns for the two age groups.
Compared to the prior work [14] which focuses on predicting from volume conductor models, the test errors from the proposed method seems to be much larger. Can the author comment on whether that suggests a major drawback of the strategy of predicting directly from T1 MRIs?
- Please rate the clarity and organization of this paper
Very Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Do you have any additional comments regarding the paper’s reproducibility?
N/A
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
The reported results make one wonder if the age group difference requires more careful treatment in the modeling step, especially considering that the older cohort comes from a Phase III clinical trial where participants might suffer from cognitive impairment. The task of generating digit twins for healthy versus pathological patients seems to be related to pseudo-health synthesis where disentanglement approaches have been studied.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making
Accept — should be accepted, independent of rebuttal (5)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The approach introduced in this paper is promising, particularly with its innovative use of deep learning frameworks and its application to real-time tDCS electric field prediction. This leads to future work with potential impact in clinical practice.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
N/A
- [Post rebuttal] Please justify your decision
N/A
Review #4
- Please describe the contribution of the paper
This study presents a novel deep learning approach for generating electrical field maps for tDCS directly from T1-weighted magnetic resonance images (MRIs). This innovation has the potential to significantly impact the field of tDCS research:
1.Individualized current dose estimation: The proposed method offers the possibility of personalized tDCS protocols by estimating the specific electrical field distribution within an individual’s brain based on their unique MRI. 2.Improved generalizability: The inclusion of special loss regularizations aims to enhance the model’s generalizability across different head shapes and electrode configurations. 3.Faster processing: Compared to physics-based methods, this deep learning approach promises significantly faster turnaround times, generating results in minutes instead of hours.
- Please list the main strengths of the paper; you should write about a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
1.High innovation potential: The concept of using deep learning for individualized tDCS electrical field modeling represents a significant leap forward in the field. 2.Focus on generalizability: The inclusion of regularization methods to improve generalizability across diverse head shapes and electrode placements strengthens the approach. 3.Large dataset: Utilizing 442 unique individual heads for training contributes to the potential robustness of the model. 4.Faster processing: The substantial reduction in processing time compared to traditional methods offers a major practical advantage.
- Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
1.Limited validation: As a proof-of-concept study, the validation process might require further exploration to establish the accuracy and reliability of the generated electrical field maps. 2.Potential for overfitting: Even with regularization, the deep learning model could still be susceptible to overfitting on the training data.
- Please rate the clarity and organization of this paper
Very Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The authors claimed to release the source code and/or dataset upon acceptance of the submission.
- Do you have any additional comments regarding the paper’s reproducibility?
None
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
1.Future validation: The authors could acknowledge the need for more extensive validation studies involving real-world tDCS applications and comparisons with established methods. 2.Error estimation: Including estimations of potential errors associated with the generated electrical field maps would provide users with a more comprehensive understanding of the model’s limitations.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making
Accept — should be accepted, independent of rebuttal (5)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Despite the limitations in validation, the high innovation potential and the clear benefits of faster processing times warrant an “Accept” recommendation. The authors can address concerns about overfitting through further validation and by providing error estimations in future work. This research offers a promising new direction for personalized tDCS interventions, and further development could significantly enhance the field.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
N/A
- [Post rebuttal] Please justify your decision
N/A
Author Feedback
N/A
Meta-Review
Meta-review not available, early accepted paper.