Cell segmentation without annotation by unsupervised domain adaptation based on cooperative self-learning

Abstract

Cell segmentation, which extracts cells from microscopic images, is essential for quantitative evaluation of cell morphology. Recently, supervised deep-learning-based models have been shown to achieve highly accurate segmentation. However, the performance of these supervised models is often degraded when the models infer unknown cell types that are not included in the train dataset. One approach to overcoming the performance degradation is acquiring new annotated data for each cell type. However, constructing datasets for all cell types is not feasible because labeling every single pixel, rather than each image, is required in the segmentation task. Learning methods that can achieve highly accurate segmentation without annotation is strongly required.

Here, we developed a cell segmentation method based on unsupervised domain adaptation with cooperative self-learning (CULPICO: Cooperative Unsupervised Learning for PIxel-wise COloring). The proposed method consists of two independent segmentation models and a mutual exchange mechanism of inference data. For the data with labels, the models are trained through supervised learning. For the data without labels, the models infer a label probability at each pixel and generate a pseudo-label as unsupervised learning. The pseudo-labels created by each model are mutually used as ground-truth in the other model. Loss function is corrected by considering pixel-level discrepancies between the label probabilities inferred by the two models. The proposed method, despite being an unsupervised learning method, can segment the unknown cell types without labels with an accuracy comparable to supervised learning models. Our method, which could solve the performance degradation problem without constructing new datasets, is expected to accelerate life science by reducing the cost of extracting quantitative biological knowledge.