Scribble2Label: Self-labeling via Consistency for Scribble-supervised Cell Segmentation

Abstract

Cell segmentation gives important findings in medical image analysis. Through cell analysis, various tasks such as cancer diagnosis, reconstruction of synaptic connectivity maps, measurement of drug response and so on could be possible. With the advent of recent advances in deep learning, more accurate and high-throughput cell segmentation has become feasible. However, deep learning-based cell segmentation faces a problem of cost and scalability for constructing dataset. Supervised-learning methods require fully annotated ground-truth labels, where there are as many as hundreds of cells. Consequently, it needs time-consuming and labor-intensive works. In this thesis, Scribble2Label, a novel weakly-supervised cell segmentation framework that exploits only a handful of scribble annotations without full segmentation labels. The core idea is to combine pseudo-labeling and label filtering to generate reliable labels from weak supervision. For this, we leverage the consistency of predictions by iteratively averaging the predictions to improve pseudo labels. The performance of Scribble2Label is demonstrated by comparing it to several state-of-the-art cell segmentation methods with various cell image modalities, including bright-field, fluorescence, and electron microscopy. Our method achieves outperformed results compared with previous related works from various data including fluorescence, histopathology, Bright-field and electron microscopy(EM). Furthermore, the prop method consistently works well in different scribble instance levels.