Utilizing explainable ML algorithms to distinguish between HCs and adolescent MDD patients
Poster No:
370
Submission Type:
Abstract Submission
Authors:
Baoshuai Zhang1,2, Baolin Wu1,2, Lei Li1,2, Qiyong Gong1,3
Institutions:
1Department of Radiology and HMRRC, West China Hospital, Sichuan University, Chengdu, China, 2Research Unit of Psychoradiology, Chinese Academy of Medical Sciences, Chengdu, China, 3Department of Radiology, West China Xiamen Hospital of Sichuan University, Xiamen, China
First Author:
Baoshuai Zhang
Department of Radiology and HMRRC, West China Hospital, Sichuan University|Research Unit of Psychoradiology, Chinese Academy of Medical Sciences
Chengdu, China|Chengdu, China
Department of Radiology and HMRRC, West China Hospital, Sichuan University|Research Unit of Psychoradiology, Chinese Academy of Medical Sciences
Chengdu, China|Chengdu, China
Co-Author(s):
Baolin Wu
Department of Radiology and HMRRC, West China Hospital, Sichuan University|Research Unit of Psychoradiology, Chinese Academy of Medical Sciences
Chengdu, China|Chengdu, China
Department of Radiology and HMRRC, West China Hospital, Sichuan University|Research Unit of Psychoradiology, Chinese Academy of Medical Sciences
Chengdu, China|Chengdu, China
Lei Li
Department of Radiology and HMRRC, West China Hospital, Sichuan University|Research Unit of Psychoradiology, Chinese Academy of Medical Sciences
Chengdu, China|Chengdu, China
Department of Radiology and HMRRC, West China Hospital, Sichuan University|Research Unit of Psychoradiology, Chinese Academy of Medical Sciences
Chengdu, China|Chengdu, China
Qiyong Gong
Department of Radiology and HMRRC, West China Hospital, Sichuan University|Department of Radiology, West China Xiamen Hospital of Sichuan University
Chengdu, China|Xiamen, China
Department of Radiology and HMRRC, West China Hospital, Sichuan University|Department of Radiology, West China Xiamen Hospital of Sichuan University
Chengdu, China|Xiamen, China
Introduction:
Major depressive disorder (MDD) tends to emerge during adolescence; however, neurobiological research in adolescents has lagged behind that in adults (Marx et al., 2023). At present, there exists a paucity of research concerning the application of machine learning (ML) classification techniques to differentiate between first-episode drug-naïve adolescent patients with major depressive disorder (MDD) and healthy controls (HCs). Furthermore, there is a substantial need for improvement in the interpretability of these classification models. This study aims to employ interpretable ML algorithms to classify these two cohorts and to identify critical classification factors, thereby advancing our understanding of the neuroimaging mechanisms underlying MDD.
Methods:
Participants:
This study was approved by the Research Ethics Committee of West China Hospital, Sichuan University. Written informed consent was obtained from all participants and their parents or legal guardians. This study included 93 adolescent MDD (aMDD) patients and 77 HCs.
MR data acquisition and image processing:
All participants underwent T1-weighted MRI on a 3.0-T Discovery MR750 scanner (GE Healthcare, Milwaukee, WI) with the following parameters: TR, 8.2 ms; TE, 3.2 ms; TI 450 ms, FOV, 256 × 256 mm2; matrix size, 256 × 256, voxel size, 1.0 × 1.0 × 1.0 mm3; with a thickness of 1.0 mm without a gap. Structural MRI scans were analyzed utilizing FreeSurfer (version 7.3.2) to derive morphometric parameters, such as surface area, cortical thickness, and volume based on the Desikan-Killiany atlas.
Network construction and analysis:
The extracted cortical thickness data was employed to construct individual morphological networks based on Jensen–Shannon divergence (Wang et al., 2016). The topological characteristics of these networks were examined using graph theoretical network analysis, facilitated by the GRETNA toolbox (version 2.0.0) in Matlab2020b (Wang et al., 2015).
ML models:
Six different ML algorithms (decision tree, extreme gradient boosting (XGBoost), multilayer perceptron (MLP) ,Elastic Net (ENet), k-nearest neighbor classification (kNN) and ridge regression) implemented by tidymodels (https://www.tidymodels.org/). Grid search or Bayesian optimization techniques was used for hyperparameter optimization and explainable SHapley Additive exPlanations (SHAP) algorithm was utilized to analyze the contribution of each of the features at the population level. The features incorporated into the ML model included cortical area, volume, thickness, as well as global and local topological metrics, following a feature screening process using the T-test. The classifier's performance was assessed using 10-fold cross-validation.
This study was approved by the Research Ethics Committee of West China Hospital, Sichuan University. Written informed consent was obtained from all participants and their parents or legal guardians. This study included 93 adolescent MDD (aMDD) patients and 77 HCs.
MR data acquisition and image processing:
All participants underwent T1-weighted MRI on a 3.0-T Discovery MR750 scanner (GE Healthcare, Milwaukee, WI) with the following parameters: TR, 8.2 ms; TE, 3.2 ms; TI 450 ms, FOV, 256 × 256 mm2; matrix size, 256 × 256, voxel size, 1.0 × 1.0 × 1.0 mm3; with a thickness of 1.0 mm without a gap. Structural MRI scans were analyzed utilizing FreeSurfer (version 7.3.2) to derive morphometric parameters, such as surface area, cortical thickness, and volume based on the Desikan-Killiany atlas.
Network construction and analysis:
The extracted cortical thickness data was employed to construct individual morphological networks based on Jensen–Shannon divergence (Wang et al., 2016). The topological characteristics of these networks were examined using graph theoretical network analysis, facilitated by the GRETNA toolbox (version 2.0.0) in Matlab2020b (Wang et al., 2015).
ML models:
Six different ML algorithms (decision tree, extreme gradient boosting (XGBoost), multilayer perceptron (MLP) ,Elastic Net (ENet), k-nearest neighbor classification (kNN) and ridge regression) implemented by tidymodels (https://www.tidymodels.org/). Grid search or Bayesian optimization techniques was used for hyperparameter optimization and explainable SHapley Additive exPlanations (SHAP) algorithm was utilized to analyze the contribution of each of the features at the population level. The features incorporated into the ML model included cortical area, volume, thickness, as well as global and local topological metrics, following a feature screening process using the T-test. The classifier's performance was assessed using 10-fold cross-validation.
Results:
There were no significant differences in gender (p = 0.551), age (p = 0.965), education level (p = 0.631) and BMI (p = 0.935) between aMDD patients and HCs (Fig. 1a). After feature screening, 38 features were finally used to build the ML model. While the decision tree algorithm exhibited overfitting and produced suboptimal classification results on the test dataset, the MLP, Elastic Net, kNN, and ridge regression algorithms demonstrated robust performance. Among these, the Elastic Net algorithm achieved the highest performance on the test dataset (AUC: 0.957 (0.926-0.989), PPV: 0.900 and NPV: 0.877). And the test AUC was 0.684 (0.517-0.852), with PPV 0.696 and NPV 0.600 (Fig. 1b, Fig. 2). Core predictors in SHAP included the degree of the left pars triangularis, the volume of the right frontal pole cortex, and the thickness of the left supramargina gyrus and the right lingual gyrus (Fig. 2).
·Figure 1. Demographic and clinical characteristics of the participants, and performance of the classification models for differentiating adolescent MDD patients and HCs.
·Figure 2. The flowchart, model performance and the impact of features on classification predictions of MDD and HCs.
Conclusions:
The MLP, Elastic Net, kNN, and ridge regression algorithms are capable of accurately differentiating between first epicode drug-naïve adolescent MDD patients and Hcs. Ultilizing SHAP method, the observed variations in gray matter patterns contribute to a deeper understanding of the potential pathophysiological mechanisms underlying MDD.
Disorders of the Nervous System:
Psychiatric (eg. Depression, Anxiety, Schizophrenia) 1
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Keywords:
Cortex
Machine Learning
STRUCTURAL MRI
Other - first-episode drug-naïve adolescent MDD; Morphological Network; Graph Theory Analysis; SHapley Additive exPlanations
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
2. Wang, J., Wang, X., Xia, M., Liao, X., Evans, A., & He, Y. (2015). GRETNA: a graph theoretical network analysis toolbox for imaging connectomics. Front Hum Neurosci, 9, 386.
3. Wang, H., Jin, X., Zhang, Y., & Wang, J. (2016). Single-subject morphological brain networks: connectivity mapping, topological characterization and test-retest reliability. Brain Behav, 6(4), e00448.