Predicting bipolar disorder using autoencoder and machine learning stacking ensembles with brain MRI
Poster No:
464
Submission Type:
Abstract Submission
Authors:
Ju-Yun Cheng1, Che-Yu Hsu2, Huai-Hsuan Tseng3,4, Po-See Chen3,4, Jun-Cheng Weng2
Institutions:
1Department of Artificial Intelligence, Chang Gung University, Taoyuan, Taiwan, 2Department of Medical Imaging and Radiological Sciences, Chang Gung University, Taoyuan, Taiwan, 3Institute of Behavioral Medicine, College of Medicine, National Cheng Kung University, Tainan, Taiwan, 4Department of Psychiatry, National Cheng Kung University Hospital, College of Medicine, National Cheng Kung University, Tainan, Taiwan
First Author:
Co-Author(s):
Che-Yu Hsu
Department of Medical Imaging and Radiological Sciences, Chang Gung University
Taoyuan, Taiwan
Department of Medical Imaging and Radiological Sciences, Chang Gung University
Taoyuan, Taiwan
Huai-Hsuan Tseng
Institute of Behavioral Medicine, College of Medicine, National Cheng Kung University|Department of Psychiatry, National Cheng Kung University Hospital, College of Medicine, National Cheng Kung University
Tainan, Taiwan|Tainan, Taiwan
Institute of Behavioral Medicine, College of Medicine, National Cheng Kung University|Department of Psychiatry, National Cheng Kung University Hospital, College of Medicine, National Cheng Kung University
Tainan, Taiwan|Tainan, Taiwan
Po-See Chen
Institute of Behavioral Medicine, College of Medicine, National Cheng Kung University|Department of Psychiatry, National Cheng Kung University Hospital, College of Medicine, National Cheng Kung University
Tainan, Taiwan|Tainan, Taiwan
Institute of Behavioral Medicine, College of Medicine, National Cheng Kung University|Department of Psychiatry, National Cheng Kung University Hospital, College of Medicine, National Cheng Kung University
Tainan, Taiwan|Tainan, Taiwan
Jun-Cheng Weng
Department of Medical Imaging and Radiological Sciences, Chang Gung University
Taoyuan, Taiwan
Department of Medical Imaging and Radiological Sciences, Chang Gung University
Taoyuan, Taiwan
Introduction:
Bipolar disorder (BD) is an affective disorder characterized by "bipolar" mood phases of mania and depression, each lasting from a few days to a few weeks. With a global prevalence of approximately 1% (Grande et al., 2016), is among the leading causes of disability worldwide.
The project is to predict the differences in brain imaging between individuals with BD and those with normal brain function. Diagnosis of BD has traditionally relied on clinical interviews and observation of symptoms, which are often insufficient for detecting variations in brain structure. To address this limitation, a predictive program utilizing an unsupervised learning approach within a stacking model has been developed. Observing which of the different machine combinations is the most effective. It is expected that this model will enhance the diagnostic process and inform treatment strategies for BD.
The project is to predict the differences in brain imaging between individuals with BD and those with normal brain function. Diagnosis of BD has traditionally relied on clinical interviews and observation of symptoms, which are often insufficient for detecting variations in brain structure. To address this limitation, a predictive program utilizing an unsupervised learning approach within a stacking model has been developed. Observing which of the different machine combinations is the most effective. It is expected that this model will enhance the diagnostic process and inform treatment strategies for BD.
Methods:
The research was conducted as a cross-sectional study, 80 patients diagnosed with BD were enrolled from the National Cheng Kung University Hospital, and 95 healthy controls (HCs) were enrolled from the community.
To overcome the imbalance between the BD and HCs, the synthetic minority oversampling technique (SMOTE) (Chawla, Bowyer et al., 2002) was applied to generate synthetic samples, ensuring a balanced dataset representation. Utilizing data augmentation such as rotation, shifting, and flipping to expand the dataset and enhance its diversity. An autoencoder extracted key features from 3D medical images via 3D convolutional layers, reducing dimensionality while retaining critical classification information (Baldi, 2012).
A stacking approach combined their outputs, with a Logistic Regression (LR) meta-learner improving prediction accuracy (Pavlyshenko, 2018). The study will adjust the base learner in the stacking model, four types of machine learning used are Extra Trees (Extremely Randomized Trees), Random Forest (RF), K-Nearest Neighbors (KNN), eXtreme Gradient Boosting (XGB) (Singh, Thakur et al., 2016). Three of them are used as a group, and all of them are used together to make a total of five combinations (Figure 1), which is intended to explore the accuracy of the model between different combinations. Cross-validation tracked training and validation losses, while boxplots, confusion matrices, classification reports, and area under curve (AUC) plots evaluated model performance.
To overcome the imbalance between the BD and HCs, the synthetic minority oversampling technique (SMOTE) (Chawla, Bowyer et al., 2002) was applied to generate synthetic samples, ensuring a balanced dataset representation. Utilizing data augmentation such as rotation, shifting, and flipping to expand the dataset and enhance its diversity. An autoencoder extracted key features from 3D medical images via 3D convolutional layers, reducing dimensionality while retaining critical classification information (Baldi, 2012).
A stacking approach combined their outputs, with a Logistic Regression (LR) meta-learner improving prediction accuracy (Pavlyshenko, 2018). The study will adjust the base learner in the stacking model, four types of machine learning used are Extra Trees (Extremely Randomized Trees), Random Forest (RF), K-Nearest Neighbors (KNN), eXtreme Gradient Boosting (XGB) (Singh, Thakur et al., 2016). Three of them are used as a group, and all of them are used together to make a total of five combinations (Figure 1), which is intended to explore the accuracy of the model between different combinations. Cross-validation tracked training and validation losses, while boxplots, confusion matrices, classification reports, and area under curve (AUC) plots evaluated model performance.
Results:
The results highlight the performance of five models across test accuracy, confusion matrices, classification reports, and AUC values. Model 1 demonstrates stable performance (test accuracy: 0.8843, AUC: 0.91) with moderate misclassification on negative samples. Model 2 underperforms with lower test accuracy (0.864) and AUC (0.88), struggling to handle negative samples effectively. Model 3 improves with better balance (test accuracy: 0.8829, AUC: 0.92), minimizing misclassification. Model 4 exhibits balanced precision and recall, with reduced false positives and negatives, though its test accuracy (0.8755) and AUC (0.88) are slightly lower. Model 5 outperforms all others, achieving the highest test accuracy (0.8971) and AUC (0.94), with low misclassification rates and superior handling of both classes. Overall, Model 5 is the most accurate and reliable, excelling across all evaluation metrics.
Conclusions:
This study demonstrates the effectiveness of using machine learning models combined with a stacking approach to distinguish individuals with bipolar disorder from healthy controls based on brain imaging data. Among the evaluated models, the stacking classifier with all components (Model 5) achieved the highest accuracy (0.8971) and AUC (0.94), demonstrating superior performance in handling both classes. These results indicate that predictive models integrating diverse machine learning algorithms may facilitate the identification of brain imaging differences, potentially contributing to improved diagnosis and treatment strategies for bipolar disorder.
Disorders of the Nervous System:
Psychiatric (eg. Depression, Anxiety, Schizophrenia) 1
Modeling and Analysis Methods:
Classification and Predictive Modeling
Diffusion MRI Modeling and Analysis 2
Novel Imaging Acquisition Methods:
Diffusion MRI
Keywords:
Affective Disorders
Design and Analysis
Informatics
Machine Learning
MRI
Psychiatric Disorders
Other - Autoencoder
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.
Chawla, N. V., Bowyer, K. W., Hall, L. O., & Kegelmeyer, W. P. (2002). SMOTE: synthetic minority over-sampling technique. Journal of artificial intelligence research, 16, 321-357.
Grande, I., Berk, M., Birmaher, B., & Vieta, E. (2016). Bipolar disorder. The Lancet, 387(10027), 1561-1572.
Pavlyshenko, B. (2018). Using stacking approaches for machine learning models. 2018 IEEE second international conference on data stream mining & processing (DSMP),
Singh, A., Thakur, N., & Sharma, A. (2016). A review of supervised machine learning algorithms. 2016 3rd international conference on computing for sustainable global development (INDIACom),