Curriculum-guided unsupervised domain adaptation enhances cross-center MRI Parkinson’s diagnosis

A curriculum-guided unsupervised domain adaptation framework improved MRI-based Parkinson’s diagnosis across centers, boosting accuracy from 65.81% to 67.95% and AUC from 0.6474 to 0.7148 on a cross-domain task. It mitigates pseudo-label noise.

• Evaluated on two independent multi-cohort Parkinson’s disease MRI datasets.
• The framework combines a curriculum learning scheduler for adaptive pseudo-label refinement and a masked consistency constraint for robust feature learning.
• Ablation studies confirmed that both innovations were crucial for the performance gains.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.