Tumor classification model training and use method, device, equipment, medium and product

By using a tumor classification model training method and adjusting parameters through feature extraction networks and binary/tri-classification networks, the problem of postoperative pathological detection lag was solved, enabling accurate preoperative prediction of MPR or PCR and providing effective treatment guidance.

CN120873844BActive Publication Date: 2026-06-02GUANGZHOU NAT LAB +2
View PDF 3 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
GUANGZHOU NAT LAB
Filing Date
2024-12-10
Publication Date
2026-06-02

AI Technical Summary

Technical Problem

In traditional techniques, postoperative pathological testing for neoadjuvant immunotherapy is delayed, making it impossible to provide effective treatment guidance before the procedure or accurately predict whether the patient will achieve MPR or PCR.

Method used

A tumor classification model training method was adopted. Feature extraction network was used to extract organ description sample feature data, and binary and tri-classification networks were used to determine the prediction probability distribution. The network parameters were adjusted by combining knowledge distillation loss and classification loss to achieve preoperative prediction of MPR or PCR.

Benefits of technology

It improves the classification ability of tumor classification models, enabling accurate prediction of whether patients will achieve PCR or MPR before surgery, and provides guidance for preoperative treatment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120873844B_ABST
    Figure CN120873844B_ABST
Patent Text Reader

Abstract

The application relates to a tumor classification model training and using method, device, equipment, medium and product. The method comprises the following steps: acquiring organ description samples of sample users; extracting sample feature data of the organ description samples based on a feature extraction network in a tumor classification model; determining a prediction probability distribution of the sample users under a two-classification network according to the sample feature data based on the two-classification network in the tumor classification model; determining a prediction probability distribution of the sample users under a three-classification network according to the sample feature data based on the three-classification network in the tumor classification model; and adjusting network parameters of the tumor classification model according to the prediction probability distribution under the two-classification network and the prediction probability distribution under the three-classification network; wherein the two-classification network is used for classification of whether MPR is reached; and the three-classification network is used for three classification of not reaching MPR, reaching MPR but not reaching PCR, or reaching PCR.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Pulmonary nodule benign and malignant classification method based on two-channel network

    CN112232433A

  • Diagnosis method and device based on convolutional neural network and multi-modal medical image

    CN113888470A

  • Target detection method and device based on neural network, electronic equipment and medium

    CN118196370A