Weakly supervised gland instance segmentation method based on point labeling

Abstract

The invention relates to a weakly supervised gland instance segmentation method based on point labeling. The method comprises the following steps: acquiring a colon histopathology image; carrying outpoint labeling on gland instances existing in the image to generate a gland point detection training sample set; establishing a gland point detection model; performing deep learning training on the gland point detection model by using the gland point detection training sample set; predicting a high confidence point in the colon histopathology image by using the gland point detection model after deep learning training, and generating a gland instance segmentation training sample set; establishing a gland instance segmentation model; performing deep learning training on the gland instance segmentation model by using the gland instance segmentation training sample set; and using the gland instance segmentation model after deep learning training to segment the gland instance of the colon histopathology image. According to the method, automatic segmentation of the gland instance in the colon histopathology image is realized, and the data annotation cost of artificial segmentation of the gland in colon cancer diagnosis is reduced.

Description

technical field [0001] The invention belongs to the field of computer vision and medical image analysis, and in particular relates to a point-labeled weakly supervised gland instance segmentation method. Background technique [0002] Colon cancer is a common malignant tumor of the digestive tract that occurs in the colon, and it is one of the cancers with the highest incidence of new cancers. Tumor cells and tissues show certain structural features different from normal cells and tissues when observed under a microscope, also called histological features. Doctors grade the cancer based on histological features to determine the patient's cancer status and treatment options. Accurately segmenting glandular instances from pathological images is a key step for pathologists to quantitatively analyze the malignancy of adenocarcinoma for diagnosis. However, manually segmenting glandular instances in pathological images is a very time-consuming task. Pathological images are staine...

AI Technical Summary

Problems solved by technology

The classification model cannot effectively distinguish the difference between glandular regions and non-glandular regions, resulting in the inability of current weakly supervised instance segmentation algorithms to be applied in glandular segmentation

Images