An efficient Predictive Accuracy-based Active Learning (PAAL) method that leverages a lightweight Accuracy Predictor and a Weighted Polling Strategy to achieve high-performance medical image segmentation with significantly reduced annotation costs.
