Uncertainty-Guided Mutual Consistency Learning for Semi-Supervised Medical Image Segmentation

Abstract

Medical image segmentation is a fundamental and critical step in many clinical approaches. Semi-supervised learning has been widely applied to medical image segmentation tasks since it alleviates the heavy burden of acquiring expertexamined annotations and takes the advantage of unlabeled data which is much easier to acquire. Although consistency learning has been proven to be an effective approach by enforcing an invariance of predictions under different distributions, existing approaches cannot make full use of region-level shape constraint and boundary-level distance information from unlabeled data. In this paper, we propose a novel uncertainty-guided mutual consistency learning framework to effectively exploit unlabeled data by integrating intra-task consistency learning from up-todate predictions for self-ensembling and cross-task consistency learning from task-level regularization to exploit geometric shape information. The framework is guided by the estimated segmentation uncertainty of models to select out relatively certain predictions for consistency learning, so as to effectively exploit more reliable information from unlabeled data. We extensively validate our proposed method on two publicly available benchmark datasets: Left Atrium Segmentation (LA) dataset and Brain Tumor Segmentation (BraTS) dataset. Experimental results demonstrate that our method achieves performance gains by leveraging unlabeled data and outperforms existing semisupervised segmentation methods.

Learning with Limited Annotations: A Survey on Deep Semi-Supervised Learning for Medical Image Segmentation

“…Zhang et al [69] propose dualtask mutual learning framework by encouraging dual-task networks to explore useful knowledge from each other. Based on dual-task framework, Zhang et al [70] utilize both segmentation task and regression task for self-ensembling and utilize estimated uncertainty to guide the mutual consistency learning and obtain further performance improvement. Chen et al [71] propose a dual-task consistency joint learning framework that encouraged the segmentation results to be consistent with the transformation of the signed distance map predictions.…”

A Feasibility Study on Deep Learning Based Brain Tumor Segmentation Using 2D Ellipse Box Areas

“…To further evaluate the proposed scheme, Table 6 shows the comparison of the dice scores from several state-of-the-art methods, as well as the method using fully annotated GT tumor areas for training (i.e., the "conventional" method). It is worth noting that the results from the methods [8,24] in Table 6 can only be used as an indication of performance because they were trained on a much larger BraTS'19 as comparing to the one using BraTS'17 [30]. Observing the results in bold fonts in Table 6, the "conventional" method resulted with the best segmented performance as 0.9001 and the proposed method as 0.8407.…”

“…However, in medical applications, such approaches are still being exploited. Zhang et al [24] proposed a semi-supervised method that exploits information from unlabeled data by estimating segmentation uncertainty in predictions, and Luo et al [25] used a dual-task deep network to predict a segmentation map and geometry-aware level set labels. Ali et al proposed the use of rectangular shape [26] and ellipse shape [27] bounding box tumor regions for tumor classification.…”

CAFS: An Attention-Based Co-Segmentation Semi-Supervised Method for Nasopharyngeal Carcinoma Segmentation

“…Lou et al [52] proposed a semi-supervised method that extends the backbone segmentation network to produce pyramidal predictions at different scales. Zhang et al [53] then use the teacher's uncertainty estimates to guide the student and perform consistent learning to uncover more information from the unlabeled data.…”