Rethinking deep active learning for medical image segmentation: A diffusion and angle-based framework

Linhao Qu, Qiuye Jin, Kexue Fu, Manning Wang, Zhijian Song

Abstract: Highlights•Introduce DifABAL, a one-shot active learning framework for medical image segmentation.•Utilize diffusion model-based autoencoders to extract features from unlabeled samples.•Develop a novel angle-based query strategy that is robust to parameter changes.•Demonstrate remarkable performance on pathology, chest X-ray, and dermatology images.