Enhancing multi-center generalization of machine learning-based depression diagnosis from resting-state fMRI

Abstract

<jats:p>Resting-state fMRI has the potential to find abnormal behavior in brain activity and to diagnose patients with depression. However, resting-state fMRI has a bias depending on the scanner site, which makes it difficult to diagnose depression at a new site. In this paper, we propose methods to improve the performance of the diagnosis of major depressive disorder (MDD) at an independent site by reducing the site bias effects using regression. For this, we used a subgroup of healthy subjects of the independent site to regress out site bias. We further improved the classification performance of patients with depression by focusing on melancholic depressive disorder. Our proposed methods would be useful to apply depression classifiers to subjects at completely new sites.</jats:p>

Evidence of shared and distinct functional and structural brain signatures in schizophrenia and autism spectrum disorder

“…The third step was further regressing out the data effect from each subject's measures that already removed the age, gender, and site effects. The processing is consistent with our previous work 29 and is also similar to some other studies 71 .…”

Multi-disciplinary fairness considerations in machine learning for clinical trials

“…Acquiring useful data for ML is complicated by noisy data collection processes and differences in formats of EHR and other clinical data [68], which could be standardized. Generalization may be improved through existing methods, including use of external validation datasets [64] or multiple data sites [179], and through identification and manipulation of covariates that impact generalizability [90]. ML may also assist in identification of dataset drift, and provide recommendations of appropriate model-updating procedures [63], which can then be resolved through other methods, such as periodic model validation and manual model re-training.…”

Inadequate dataset learning for major depressive disorder MRI semantic classification

“…Transfer learning is an essential tool in machine learning that solves the fundamental problem of insufficient training data and can be applied to many areas that are difficult to improve due to inadequate training data [16,17]. Researchers have conducted fMRI studies on depression [18,19]. Multiple brain regions are abnormally active in patients on task-state fMRI.…”