Cervical cancer histopathology image analysis method and device based on depth learning

Abstract

The invention discloses a cervical cancer histopathology image analysis method and device based on depth learning. The method comprises the following steps: acquiring cervical cancer histopathology images and setting image labels for each image; training two convolution neural networks respectively based on the images used for training, and obtaining the two trained convolution neural networks; fixing parameters of the two trained convolutional neural networks, and training the all-connected layer one based on the images used for training, so that the trained all-connected layer one is obtained. The image to be tested is input into the trained classifier, and the two convolution neural networks extract feature vectors from the image respectively. The output feature vectors f1 and f2 are spliced together and input to the full connection layer 1, and the output feature vector f3. The classification result is determined by the element with the largest value in the feature vector f3 The invention can automatically classify and identify the differentiation degree of the original histopathological slice micrograph collected by the doctor and assist the doctor in diagnosis.

Description

technical field [0001] The present invention relates to the medical field, in particular to a method and device for analyzing cervical cancer histopathological images based on deep learning. Background technique [0002] Cervical cancer is one of the common gynecological malignancies. For now, research on computer-aided diagnosis of histopathological microscopic images of cervical cancer mainly focuses on the classification of cervical histopathological images using classical image feature extraction methods and machine learning classification methods. Image segmentation and pathological abnormality screening, few studies have conducted computer-aided diagnosis research on the degree of differentiation of cervical cancer histopathological images. [0003] The prior art is based on using computer vision methods to extract features from histopathological images of only a single cervical cancer cell, and then using traditional machine learning methods to further classify the ex...

AI Technical Summary

Problems solved by technology

[0009] However, this method can only select a single cell histopathological image with low noise and high definition for processing, and cannot directly process the complete histopathological microsection image, and needs to manually extract the single cell image; this method uses 27 artificially designed morphological features are used as the input value of machine learning. This method has poor transferability and is prone to overfitting in learning; and it can only classify the cell type of a single cell, and cannot diagnose the patient's cancer globally. The degree of differentiation, thus assisting in judging the patient's cancer condition

Images