MSRCN: Multi-task Learning Network for Cell Segmentation and Regression Counting

Abstract

Accurate cell segmentation and counting play an important role in medical diagnosis. However, the size and shape of cells are varied largely, and the presence of overlapping cells complicates cell counting. Recent studies have shown that multi-task learning methods perform well in deep learning. In specific, we design Multi-task Segmentation Regression Counting Network (MSRCN). For cell segmentation, a multi-scale attention mechanism module is designed to suppress irrelevant regions and learns salient features for a specific task. For cell counting, a regression model is utilized to learn a mapping from cell feature information to target counts. The proposed MSRCN model is analyzed and compared with other states of the art cell segmentation methods and cell counting methods. MSRCN outperforms these methods in all evaluation metrics. The Dice similarity coefficient, root mean square error, and mean absolute error of the proposed method is 0.9316, 2.1215, and 1.5927, respectively. The experiments results show that the proposed method not only improves the functioning of cell segmentation, but also outperforms direct regression counting methods in terms of cell counting.