1 August 2021 in Medical Image Analysis
17 May 2021 in IEEE Journal of Biomedical and Health Informatics
Automatic and accurate detection of anatomical landmarks is an essential operation in medical image analysis with a multitude of applications. Recent deep learning methods have improved results by directly encoding the appearance of the captured anatomy with the likelihood maps (i.e., heatmaps). However, most current solutions overlook another essence of heatmap regression, the objective metric for regressing target heatmaps and rely on hand-crafted heuristics to set the target precision, thus being usually cumbersome and task-specific. In this paper, we propose a novel learning-to-learn framework for landmark detection to optimize the neural network and the target precision simultaneously. The pivot of this work is to leverage the reinforcement learning (RL) framework to search objective metrics for regressing multiple heatmaps dynamically during the training process, thus avoiding setting problem-specific target precision. We also introduce an early-stop strategy for active termination of the RL agent's interaction that adapts the optimal precision for separate targets considering exploration-exploitation tradeoffs. This approach shows better stability in training and improved localization accuracy in inference. Extensive experimental results on two different applications of landmark localization: 1) our in-house prenatal ultrasound (US) dataset and 2) the publicly available dataset of cephalometric X-Ray landmark detection, demonstrate the effectiveness of our proposed method. Our proposed framework is general and shows the potential to improve the efficiency of anatomical landmark detection.
21 October 2023 in Medical Image Analysis
23 September 2020 in IEEE Access
Deep neural networks are very compelling for medical image segmentation. However, deep models often suffer from notable performance drops in real clinical settings due to the complex appearance shift in daily scannings. Domain adaptation partially addresses the problem between imaging domains. However, it heavily depends on the expensive re-collection and re-training for domain-specific datasets and thus is not applicable to domain-agnostic images. In this paper, we propose a case adaptation strategy aiming to bridge the segmentation performance gap on domain-agnostic images. Our contribution is three-fold. First, we design a general self-supervised learning framework for case adaptation , which exploits its predictions as supervision to drive the adaptation. Without extra annotations and any burden on model complexity, the framework enables trained deep models at-hand to directly segment domain-agnostic testing images. Second, we propose a novel Evolving Shape Prior (ESP) which recursively introduces strong shape knowledge into networks and evolves with the adaptation procedure to provide adaptive supervision. ESP can stabilize self-supervised learning and guide it to move towards model convergence. Third, we perform extensive experiments on 10 datasets with different levels of difficulty and typical appearance shifts blended, proving our framework is a promising solution in reducing segmentation performance degradation. Through this work, we investigate the feasibility of case adaptation as a general strategy in enhancing the robustness of deep segmentation networks, with comprehensive analyses proving its efficacy and efficiency.