Related Experiment Video
Updated: Feb 21, 2026
12:28
Abbiategrasso Brain Bank Protocol for Collecting, Processing and Characterizing Aging Brains
Published on: June 3, 2020
18.3K
Deep Learning-Based Structural Brain Age Estimation in Bipolar Disorder and Schizophrenia: A Single-Site Pilot Study.
Akila Weerasekera1,2,3, Shuqin Zhou1,2,3, Chao Wang1,2,3
1Psychotic Disorders Division, McLean Hospital, Belmont, Massachusetts, USA.
Human Brain Mapping
|February 19, 2026
Summary
Schizophrenia (SZ) and bipolar disorder (BD) show accelerated brain aging (Brain-PAD) early in adulthood, with greater variability than healthy controls (HC). Frontotemporal regions are key drivers of these brain age predictions.
Area of Science:
- Neuroscience
- Psychiatry
- Radiology
Background:
- Accelerated brain aging is linked to severe mental illnesses like schizophrenia (SZ) and bipolar disorder (BD).
- Brain-PAD, a neuroimaging biomarker, shows potential but requires further understanding of its development and variability in SZ and BD.
- Existing research lacks detailed analysis of regional brain aging patterns in these conditions.
Purpose of the Study:
- To investigate the developmental trajectory and regional drivers of Brain-PAD in SZ and BD using a deep learning approach.
- To assess within-group variability and identify brain regions contributing to age prediction in SZ and BD patients.
- To compare Brain-PAD patterns between SZ, BD, and healthy controls (HC).
Main Methods:
- A 3D convolutional neural network (3D-CNN) was trained on structural MRI data from HC (n=155) and applied to independent SZ (n=161) and BD (n=122) cohorts.
- Brain-PAD was calculated as predicted brain age minus chronological age.
- Gradient-weighted Class Activation Mapping (Grad-CAM) identified regional contributions; statistical analyses examined age-by-group interactions and heterogeneity.
Main Results:
- The 3D-CNN model demonstrated high accuracy in HC (MAE=3.05 years) but reduced accuracy in BD (MAE=8.86 years) and SZ (MAE=9.01 years).
- Mean Brain-PAD was significantly elevated in SZ (+6.7 years) and BD (+4.2 years) compared to HC (+0.7 years), particularly at younger ages.
- SZ and BD showed increased Brain-PAD heterogeneity and divergence from HC trajectories after age 40. Frontotemporal regions were key predictors of brain age across groups.
Conclusions:
- SZ and BD exhibit accelerated apparent brain aging (Brain-PAD) early in adulthood, with greater heterogeneity than HC.
- Frontotemporal regions are critical for brain age prediction, indicating sensitivity to age-related structural changes.
- Brain-PAD serves as a group-level marker of apparent brain aging, warranting longitudinal investigation, especially regarding midlife divergence.
