Chest digital X-ray image processing method and system for pneumoconiosis

By segmenting and data augmenting the digital X-ray image, combining the deep backbone classification network model and multi-supervised classification loss function, the student network is trained, and overfitting and low classification accuracy problems caused by limited data volume and weak lesion characteristics in the existing technology are solved, achieving higher classification accuracy and robustness.

CN119693305AActive Publication Date: 2025-03-25SHENZHEN YUDE TECHNOLOGY DEVELOPMENT CO LTD
11 Cites 1 Cited by

Patent Information

Application Number
CN202411665210.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-11-20
Publication Date
2025-03-25
Estimated Expiration
2044-11-20

AI Technical Summary

Technical Problem

In the face of limited data volume and weak and difficult to identify lesion characteristics, existing deep learning technologies are prone to overfitting digital X-ray image classification and low classification accuracy.

Method used

A chest digital X-ray image processing method for pneumoconiosis is adopted. By segmenting and data augmenting the lung area of ​​the digital X-ray image, a global and local views are generated, and a teacher and student feature extraction network is constructed by combining the deep backbone classification network model. The student network is trained using a multi-supervised classification loss function to reduce overfitting.

Benefits of technology

It effectively improves the accuracy of digital X-ray image classification, reduces the overfitting phenomenon in the case of scarcity of data, and enhances the robustness and reliability of the model.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses a pneumoconiosis-oriented chest digital X-ray image processing method and system, and the method comprises the steps: extracting a complete rectangular region image with lung information in a digital X-ray image, and dividing the rectangular region image; for the training set, a random enhancement strategy is applied to enhance each sample, and two global views and six local views are generated; constructing a teacher and student feature extraction network, respectively inputting the two global views and the six local views into the teacher and student feature extraction network, obtaining depth features, outputting the depth features, respectively sending the depth features to a feature mapping layer and a linear classification layer, and obtaining a mapping representation and a classification result corresponding to each view; the student network is trained in combination with a loss function, and the teacher network is obtained by using an index moving average strategy; after each round of training is finished, inputting the verification set into the teacher network, and evaluating the classification performance of the verification set; and storing the teacher model weight with the best classification performance of the verification set, and inputting the test set into the teacher network to obtain a final test result.
Need to check novelty before this filing date? Find Prior Art