Multi-label classification of gynecological secretion images

Abstract

The invention discloses a multi-label classification method for gynecological secretion images, which comprises the following steps: 1, feature extraction, comprising constructing a training databaseand establishing a network structure for model training; 2, model training, comprising train gynecological secretion images and corresponding label images in a training database through a full convolution network to obtain a required sample model; 3, correcting, testing that gynecological secretion image by a sample model to obtain a predictive label map for testing, evaluating the predictive label map, and judging whether the testing result meets the requirements, if so, taking the sample model as a sample model for predicting; Otherwise, returning to the step 2, modifying each parameter fortraining and repeating the step 3 until the test result meets the requirement; 4, predicting, inputting the gynecological secretion image for detection into a sample model for predicting, and obtaining a preliminary predicted image. Compared with the prior art, the classification time is saved.

Description

technical field [0001] The invention relates to the medical field, in particular to a multi-label classification method for gynecological secretion images. Background technique [0002] At present, when the laboratory department of the hospital conducts gynecological examinations, it is necessary to classify different cells in vaginal secretions, that is, the classification of multi-label images. The most commonly used method now is manual microscopy, and the existing non-manual methods on the market cannot classify such multi-label images, and can only identify and classify one of the cells separately; this will prolong the classification time , and the existing multi-label classification algorithm is not suitable for the classification of gynecological secretion cell images. [0003] How to perform multi-label classification on gynecological secretion images through technical means is a topic worthy of research. Contents of the invention [0004] The purpose of the pre...

AI Technical Summary

Problems solved by technology

The most commonly used method now is manual microscopy, and the existing non-manual methods on the market cannot classify such multi-label images, and can only identify and classify one of the cells separately; this will prolong the classification time , and the existing multi-label classification algorithm is not suitable for the classification of gynecological secretion cell images

Images