A method for localization of pathological tissue slice scanning based on clustering and segmentation

Abstract

The invention discloses a scanning region positioning method for pathological tissue slices based on clustering and segmentation, which comprises the following steps: performing image pre-cutting on slide images; performing image noise reduction processing on the images by using Gaussian filtering; and performing image processing on the images by using a grayscale world algorithm. Color correction processing; convert the RGB space of the image into Lab space to separate red and blue; use a clustering algorithm to classify the pixels of the background and tissue areas of the image to achieve clustering and segmentation of the image; Morphological processing to remove the noise region; remove the high-light region of the image, and use a rectangular frame to enclose the result of the tissue region segmentation with a minimum circumscribed rectangle. The present invention locates the pathological tissue area of ​​the collected slide image without any manual labeling of the slide image, and at the same time can give accurate results under the condition of changing illumination, has better robustness, and realizes fast and efficient determination The location of the scan area.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method for locating scanning regions of pathological tissue slices based on clustering and segmentation. Background technique [0002] At present, the image processing of pathological tissue slices mainly adopts the method based on machine learning, and there are two main methods based on machine learning: [0003] In the first one, the significance of red and blue in the Lab color space and the Frangi filter are used to enhance the slide preview image to highlight the tissue area, and a threshold-based method is used to initially segment the tissue area; the slide preview image The noise, bubbles and other interferences in the threshold segmentation result in many misclassified areas, and then collect the regional features respectively, and then manually mark whether different areas belong to the organization and train the SVM model, and finally use the trained model ...

AI Technical Summary

Problems solved by technology

[0003] In the first one, the significance of red and blue in the Lab color space and the Frangi filter are used to enhance the slide preview image to highlight the tissue area, and a threshold-based method is used to initially segment the tissue area; the slide preview image The noise, bubbles and other interferences in the threshold segmentation result in many misclassified areas, and then collect the regional features respectively, and then manually mark whether different areas belong to the organization and train the SVM model, and finally use the trained model to analyze the initial segmentation results. Correction of the area; this method requires manual labeling of a large amount of data to train the model, which is time-consuming and laborious;

[0004] The second is to segment the preview image based on the OSTU algorithm that automatically obtains the optimal segmentation threshold, and then use the method of morphological processing to remove the noise area in the segmentation result; this method is simple in design and high in execution efficiency, but It has high requirements on the illumination of slides and the production of slides, and is not very robust

In general, the magnification is larger when scanning, and the number of images after scanning is larger, which leads to longer scanning time

Images