Multimodal brain imaging-based classification of functional constipation subtypes using machine learning

Functional constipation (FC) is a common gastrointestinal condition often accompanied by anxiety and depression status (FCAD). Gastrointestinal symptoms in FCAD patients are not fully resolved with medication, and the responses to treatment can differ significantly from those of patients without anxiety and depression (FCNAD). Given the distinct effects of FC and mental status on brain function and structure, we hypothesize that these brain differences could serve as imaging features to differentiate FCAD and FCNAD. Patients with FC (N = 187) underwent structural magnetic resonance imaging, diffusion tensor imaging scans, and completed self-reported assessments of depression and anxiety. The current study first identified FCNAD and FCAD patients with high- and low-confidence labels based on self-reported ratings. Brain structural imaging features were subsequently extracted to train and refine the classification model using a stagewise training approach. The classification model achieved an average accuracy of 89.23% during the cross-validation, and the predicted probability of FCAD was significantly correlated with mental ratings and gastrointestinal symptoms. The top 30 imaging features contributing to classification were primarily located in brain regions involved in emotional processing (temporal pole, amygdala, orbitofrontal cortex), somatosensory (insula), and motor control (corticospinal tract, inferior cerebellar peduncle). These findings highlight potential brain imaging features distinguishing FCAD and FCNAD, which may provide the incremental value over standard behavioral assessments for future personalized diagnosis and treatment strategies.