A New Semi-Supervised Deep Learning Method for Psychiatric Disorders Diagnosis
Poster No:
547
Submission Type:
Abstract Submission
Authors:
Yuhui Du1, Chuanfu Han1, Ju Niu1, Fulin Wu1, Theo van Erp2, Godfrey Pearlson3, Peter Kochunov4, Vince Calhoun5
Institutions:
1Shanxi University, Taiyuan, Shanxi, 2University of California, Irvine, CA, 3Yale University, New Haven, CT, 4University of Maryland School of Medicine, Baltimore, MD, 5GSU/GATech/Emory, Atlanta, GA
First Author:
Co-Author(s):
Introduction:
The diagnosis of mental disorders largely relies on clinical indicators that are prone to subjectivity(David & Deeley, 2024; Woolway et al., 2024). Using neuroimaging data, classifying different disorders often requires large-size high-quality datasets with accurate class labels, and clustering subjects for exploring biotypes typically ignores the use of diagnosis labels (Du et al., 2024; Jahanshad et al., 2024). Here, we propose a new semi-supervised deep learning method for psychiatric disorders diagnosis. The method utilizes a dual learning strategy to optimize classification and clustering losses by leveraging both limited labeled data and abundant unlabeled data, thereby improving model performance.
Methods:
Fig. 1 shows the framework of our method consists of an iterative process with three steps. In Step 1, both initial labeled and high-quality pseudo-labeled samples identified by our method are used for the supervised training of the neural network, guided by a classification loss. In Step 2, the neural network undergoes further training, guided by a clustering loss obtained by the semi-supervised clustering procedure. In Step 3, high-quality pseudo-labeled samples are identified by evaluating the classification-clustering consistency, and then combined with the original labeled samples for the next Step 1.
We utilized large-sample resting-state fMRI data from 708 healthy controls (HCs) and 537 SZ patients across four datasets for examining the classification ability using a leave-one-dataset-out cross-validation strategy. Considering the shared characteristics between bipolar disorder (BP) and schizophrenia (SZ), we also evaluated the clustering capability using 107 BP patients with psychosis, 107 SZ patients, and 107 HCs to identify transdiagnostic categories (called Biotypes), and further analyzed the group differences and unique brain functional changes compared to the HC group.
We utilized large-sample resting-state fMRI data from 708 healthy controls (HCs) and 537 SZ patients across four datasets for examining the classification ability using a leave-one-dataset-out cross-validation strategy. Considering the shared characteristics between bipolar disorder (BP) and schizophrenia (SZ), we also evaluated the clustering capability using 107 BP patients with psychosis, 107 SZ patients, and 107 HCs to identify transdiagnostic categories (called Biotypes), and further analyzed the group differences and unique brain functional changes compared to the HC group.
Results:
Our results support that even with a few labeled samples, our semi-supervised method obtained relatively high accuracy in distinguishing SZ from HC and outperformed the traditional classification methods relying on the same amounts of labeled data. As shown in Fig. 2, group differences between the two biotypes as well as between HC and each biotype are more pronounced than that between BP and SZ, HC and BP, and HC and SZ.
Fig. 2(A) illustrates that the functional network connectivity (FNC) differences between Biotype A and Biotype B, indicated by T-values, are more distinct than those between the traditional diagnostic groups (BP and SZ). Figure 2(B) shows the mean FNC strengths across different groups, revealing more pronounced distinctions between Biotype A and Biotype B than between BP and SZ. Compared to HC, the brain alterations in these biotypes are more substantial, highlighting their biological relevance. Relative to HC, changes in Biotype A mainly occur in the FNCs between SC and visual (VI) domains, and between SM and CB domains, while changes in Biotype B involve the FNCs between SM and VI domains, and between cognitive-control and CB domains. The significant differences between Biotype A and Biotype B are primarily observed in the FNCs between sub-cortical (SC) and sensorimotor (SM) domains, between auditory (AU) and default-mode (DM) domains, between SM and cerebellar (CB) domains, and within the CB itself.
Fig. 2(A) illustrates that the functional network connectivity (FNC) differences between Biotype A and Biotype B, indicated by T-values, are more distinct than those between the traditional diagnostic groups (BP and SZ). Figure 2(B) shows the mean FNC strengths across different groups, revealing more pronounced distinctions between Biotype A and Biotype B than between BP and SZ. Compared to HC, the brain alterations in these biotypes are more substantial, highlighting their biological relevance. Relative to HC, changes in Biotype A mainly occur in the FNCs between SC and visual (VI) domains, and between SM and CB domains, while changes in Biotype B involve the FNCs between SM and VI domains, and between cognitive-control and CB domains. The significant differences between Biotype A and Biotype B are primarily observed in the FNCs between sub-cortical (SC) and sensorimotor (SM) domains, between auditory (AU) and default-mode (DM) domains, between SM and cerebellar (CB) domains, and within the CB itself.
Conclusions:
Our method demonstrated its ability to effectively utilize limited labeled data and thoroughly mine the potential information in unlabeled data, thereby significantly enhancing the model's learning capabilities and predictive accuracy for psychiatric disorders.
Disorders of the Nervous System:
Psychiatric (eg. Depression, Anxiety, Schizophrenia) 1
Modeling and Analysis Methods:
Classification and Predictive Modeling
fMRI Connectivity and Network Modeling 2
Methods Development
Keywords:
Computing
FUNCTIONAL MRI
Machine Learning
Psychiatric Disorders
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:
Provide references using APA citation style.
Du, Y., Niu, J., Xing, Y., Li, B., & Calhoun, V. D. (2024). Neuroimage Analysis Methods and Artificial Intelligence Techniques for Reliable Biomarkers and Accurate Diagnosis of Schizophrenia: Achievements Made by Chinese Scholars Around the Past Decade. Schizophrenia Bulletin, sbae110.
Jahanshad, N., Lenzini, P., & Bijsterbosch, J. (2024). Current best practices and future opportunities for reproducible findings using large-scale neuroimaging in psychiatry. Neuropsychopharmacology, 1-15.
Woolway, G. E., Legge, S. E., Lynham, A. J., Smart, S. E., Hubbard, L., Daniel, E. R., Pardiñas, A. F., Escott-Price, V., O’Donovan, M. C., & Owen, M. J. (2024). Assessing the validity of a self-reported clinical diagnosis of schizophrenia. Schizophrenia, 10(1), 99.