Weakly-Supervised Cross-Domain Adaptation for Endoscopic Lesions Segmentation

Abstract

Weakly-supervised learning has attracted growing research attention on medical lesions segmentation due to significant saving in pixel-level annotation cost. However, 1) most existing methods require effective prior and constraints to explore the intrinsic lesions characterization, which only generates incorrect and rough prediction; 2) they neglect the underlying semantic dependencies among weakly-labeled target enteroscopy diseases and fully-annotated source gastroscope lesions, while forcefully utilizing untransferable dependencies leads to the negative performance. To tackle above issues, we propose a new weakly-supervised lesions transfer framework, which can not only explore transferable domain-invariant knowledge across different datasets, but also prevent the negative transfer of untransferable representations. Specifically, a Wasserstein quantified transferability framework is developed to highlight widerange transferable contextual dependencies, while neglecting the irrelevant semantic characterizations. Moreover, a novel selfsupervised pseudo label generator is designed to equally provide confident pseudo pixel labels for both hard-to-transfer and easyto-transfer target samples. It inhibits the enormous deviation of false pseudo pixel labels under the self-supervision manner. Afterwards, dynamically-searched feature centroids are aligned to narrow category-wise distribution shift. Comprehensive theoretical analysis and experiments show the superiority of our model on the endoscopic dataset and several public datasets.

Semi-Supervised Heterogeneous Domain Adaptation: Theory and Algorithms

“…In sweeping experiments with representative baselines and 104 SsHeDA tasks, the accuracy of JMEA outstrips that of all baselines, and KHDA outperforms all nonneural network baselines. In the future, we will consider HeDA in semantic segmentation tasks [79], [80]. Inspired by Dong et al [81], [82] that develop a novel perspective to distinguish transferable or untransferable representations across domains, we will develop a novel learning theory in semantic segmentation tasks to quantify transferability across heterogeneous domains.…”

Cross-Site Severity Assessment of COVID-19 From CT Images via Domain Adaptation

“…Weakly-supervised DA methods aim to tackle the problem where the target domain has some weakly-labeled data. For example, Dong et al [23] assume that there are some image-level labels but pixel-level ones on the target domain, and propose a quantified transferability mechanism for endoscopic lesions segmentation. Semi-supervised DA methods assume that there is a little labeled data on the target domain.…”

CurSeg: A pavement crack detector based on a deep hierarchical feature learning segmentation framework

“…However, these unsupervised algorithms can only adapt to some specific noise, and the effect is not good under different illumination conditions and crack images with complex backgrounds. Fortunately, with the development of convolutional neural networks (CNNs) and deep learning [25,26], more robust features can be learned to represent cracks more accurately than traditional methods.…”