Hematoxylin-eosin staining pathological image segmentation method based on unsupervised deep learning

Abstract

The invention discloses a hematoxylin-eosin staining pathological image segmentation method based on unsupervised deep learning, and the method comprises the steps: carrying out the preprocessing andfeature extraction after an HE staining pathological image is obtained, dividing pixels into five types through Kmeans clustering, using a given training sample full-automatic grabbing strategy for carrying out traversal grabbing on category label images obtained through clustering to obtain reliable training samples, then using a training set for training a semantic segmentation model Unet, and designing different training strategies before, in the middle and after training; segmenting the to-be-segmented image into the size conforming to model input in an overlapping manner, putting the to-be-segmented image into the trained model to obtain a prediction result, and splicing the prediction result to obtain a segmentation result of the cell nucleus foreground; and finally, carrying out accurate kernel boundary segmentation on the cell nucleus part by using a hybrid watershed segmentation method to obtain a complete segmentation result. According to the method, the high efficiency of unsupervised learning and the high precision of deep learning are organically combined, and the precision and the efficiency of cell nucleus region segmentation in a pathological image segmentation taskare remarkably improved.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a hematoxylin-eosin staining pathological image segmentation method based on unsupervised deep learning. Background technique [0002] HE staining pathological image segmentation technology is the process of accurately segmenting the nucleus area and the boundary between nuclei in the image. The main basis of pathological image segmentation technology is to use some special global feature distribution in pathological images, such as color distribution or cell shape distribution, etc., according to the difference between the color characteristics or morphological characteristics of the nucleus and the background, the HE staining histopathological image The nuclei region is isolated from the background. [0003] HE staining pathological image segmentation technology is one of the most practical technologies in medical image processing. In HE staining histopathological images, the s...

AI Technical Summary

Problems solved by technology

[0005] The limitation of the classic machine learning method is that the dyeing quality of HE dyes is easily affected by external factors, and there are large differences in the dyeing results.

Moreover, it is difficult to find a clear boundary between the nucleus and the background after staining, and the diversity of nucleus shapes also makes it difficult to establish a stable shape model in nucleus detection and segmentation, and the generalization ability is poor; the limitations of the deep learning-based nucleus segmentation method The main reason is that a large number of artificially labeled samples are needed to train the model to achieve the required generalization ability, and the manually labeled samples are cumbersome and time-consuming, and are easily affected by subjective factors.

Images