Related Experiment Video
Updated: May 27, 2025
05:33
Three-Dimensional Shape Modeling and Analysis of Brain Structures
Published on: November 14, 2019
7.0K
Classification of Major Depressive Disorder Using Vertex-Wise Brain Sulcal Depth, Curvature, and Thickness with a
Roberto Goya-Maldonado1, Tracy Erwin-Grabner1, Ling-Li Zeng2,3
1Laboratory of Systems Neuroscience and Imaging in Psychiatry (SNIP-Lab), Department of Psychiatry and Psychotherapy, University Medical Center Göttingen (UMG), Georg-August University, Göttingen, Germany.
Arxiv
|February 20, 2025
Summary
Deep learning models and SVMs failed to differentiate major depressive disorder (MDD) patients from healthy controls using brain imaging data. Current neuroimaging features and classifiers are insufficient for reliable MDD diagnosis.
Area of Science:
- Neuroimaging
- Psychiatric Disorders
- Machine Learning
Background:
- Major Depressive Disorder (MDD) affects millions globally, with ongoing debate regarding brain morphology links.
- Previous machine learning attempts using segmented cortical features for MDD classification showed low accuracy.
- Deep learning offers potential for identifying complex patterns in neuroimaging data for MDD biomarkers.
Purpose of the Study:
- To evaluate the effectiveness of deep learning (DenseNet) and Support Vector Machine (SVM) classifiers in differentiating MDD patients from healthy controls (HC).
- To test the hypothesis that integrating vertex-wise cortical features improves classification performance.
- To analyze a large, multi-site dataset to ensure generalizable results.
Main Methods:
- Utilized a large, multi-site dataset (N=7,012; 2,772 MDD, 4,240 HC) from the ENIGMA-MDD working group.
- Applied DenseNet and SVM classifiers to vertex-wise cortical features.
- Used ComBat harmonization to address multi-site data effects.
Main Results:
- Both DenseNet (51%) and SVM (53%) showed near-chance balanced accuracy when tested on unseen sites.
- Slightly improved performance (DenseNet: 58%, SVM: 55%) was observed when cross-validation included subjects from all sites, suggesting site effects.
- The integration of vertex-wise morphometric features did not yield effective MDD classification.
Conclusions:
- Current vertex-wise morphometric features and non-linear classifiers are insufficient for differentiating MDD from HC.
- MDD classification using the tested combination of features and classifiers is currently unfeasible.
- Future research should explore integrating information from other MRI modalities (fMRI, DWI) for improved diagnostic performance.
