Abstract
Universal lesion detection (ULD) has great value in clinical practice as it can detect various lesions across multiple organs. Deep learning-based detectors have great potential but require high-quality annotated training data. In practice, due to cost, expertise requirements, and the diverse nature of lesions, incomplete annotations are often encountered. Directly training ULD detectors under this condition can yield suboptimal results. Leading pseudo-label methods rely on a dynamic lesion-mining mechanism operating at the mini-batch level to address the issue of incomplete annotations. However, the quality of mined lesions in this approach is inconsistent across different iterations, potentially limiting performance enhancement. Inspired by the observation that deep models learn concepts with increasing complexity, we propose an innovative exploratory-training-based ULD (ET-ULD) method to assess the reliability of mined lesions over time. Specifically, we employ a teacher-student detection model, the teacher model is used to mine suspicious lesions, which are combined with incomplete annotations to train the student model. On top of that, we design a bounding-box bank to record the mining timestamps. Each image is trained in several rounds, allowing us to get a sequence of timestamps for the mined lesions. If a mined lesion consistently appears in the timestamp sequence, it is likely to be a true lesion, otherwise, it may just be a noise. This serves as a crucial criterion for selecting reliable mined lesions for subsequent retraining. Our experimental results confirm the effectiveness of ET-ULD, showcasing its ability to surpass existing state-of-the-art methods on two distinct lesion image datasets. Notably, on the DeepLesion dataset, ET-ULD achieved a significant enhancement, outperforming the previous leading method by 5.4% in Average Precision (AP), thus demonstrating its superior performance.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have