Picture screening method and system for esophageal cancer model training and storage medium

Abstract

The invention discloses a picture screening method and system for esophageal cancer model training, and a storage medium. The method comprises the following steps: inputting a to-be-screened static picture; clustering the static pictures by adopting a clustering algorithm according to the characteristics of the static pictures to obtain a plurality of types of static pictures; and screening the static pictures in each cluster by adopting a distance function to obtain static pictures with low similarity as training samples for establishing an esophageal cancer recognition model. According to the method, when the static pictures are input, a larger sample size can be allowed to be adopted to solve the problem of poor model generalization ability, meanwhile, the static pictures with large samples are clustered through a clustering algorithm, then the static pictures with low similarity in each cluster are screened through a distance function, and the static pictures with high similarity are obtained. And finally, on the premise that the sample coverage rate is not obviously influenced, the transformation from a large sample to a small sample is realized, and an esophageal squamous carcinoma lesion picture suitable for training and identifying narrow-band imaging is obtained.

Description

technical field [0001] The invention relates to the field of intelligent medical technology, in particular to a method, system and storage medium for screening images used for esophageal cancer model training. Background technique [0002] Esophageal cancer is one of the most common malignant tumors of the digestive tract in the world, and its pathological types mainly include esophageal squamous cell carcinoma (ESCC) and esophageal adenocarcinoma (EADC). Upper gastrointestinal endoscopy combined with histopathology is the gold standard for the diagnosis of esophageal squamous cell carcinoma. For difficult-to-find lesions, chromoendoscopy and electronic chromoendoscopy are mainly used to find them, and then targeted biopsy for diagnosis through histopathology. [0003] With the development of image recognition technology, by screening and classifying lesion pictures, training and deep learning the characteristics of lesion and non-lesion pictures to establish a deep learnin...

AI Technical Summary

Problems solved by technology

However, for the sample data to completely cover the sample space, the amount of data required will increase exponentially

The large-sample training model not only increases the amount of calculations, but also has a large difference in the amount of data of different categories. If the balance of samples is not well controlled, the generalization ability of the model will also be poor. At the same time, within the sample data, various samples such as negative samples and positive samples interact with each other. Interference leads to low sensitivity and specificity of the model, causing model training to fail

Images