Fine-grained cancer subtype classification method

Abstract

The invention discloses a fine-grained cancer subtype classification method. The method comprises the following steps: step 1, obtaining a cell nucleus segmentation and classification result in a pathological picture; step 2, extracting an instance patch according to the cell nucleus segmentation and classification prediction result; step 3, performing instance patch convolution feature extraction; and step 4, generating picture-level features of the used instance patch through a Transform model to complete classification of cancer subtypes. The method can assist pathologists to classify cancer subtypes, and the working efficiency of the doctors is improved.

Description

technical field [0001] The invention belongs to the field of intelligent pathological classification, and in particular relates to a fine-grained cancer subtype classification method. Background technique [0002] With the development of digital pathology technology, tissue slices are stored in the form of digital images, which makes it possible to automatically identify pathological patterns, and cancer subtypes, as a basic task of digital pathology analysis, attract a large number of researchers. At present, many CNN-based models have been proposed for the classification of different cancers in different tissues. These methods mainly rely on extracting some specific histological features that can be easily extracted by CNN to identify cancer subtypes, such as the structural features of tumors, but for For some cancers that require more fine-grained features for classification, the traditional model based on CNN cannot complete the classification of cancer subtypes well, su...

AI Technical Summary

Problems solved by technology

Traditional CNN-based models cannot capture these fine-grained features well, and cannot perform cancer subtype classification based on fine-grained features

In addition, there are some methods to manually construct cell-level features through the prediction results of a segmentation and classification of the nucleus. Although these methods can extract fine-grained cell-level information, they cannot obtain cell-level features and these methods are aimed at specific Tasks are not universally applicable

Images