Nasopharyngeal carcinoma distant metastasis risk prediction method and system based on cross-modal Transform and depth consistency loss

The method of predicting distant metastasis risk of nasopharyngeal carcinoma using cross-modal Transformer and depth consistency loss solves the problems of insufficient utilization of multimodal imaging information and risk ranking, achieves more accurate risk assessment and early identification of high-risk patients, and enhances the clinical application value of the prediction model.

CN122455320APending Publication Date: 2026-07-24SOUTH CHINA AGRICULTURAL UNIVERSITY +1
0 Cites 0 Cited by

Patent Information

Application Number
CN202610351328.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-03-21
Publication Date
2026-07-24

AI Technical Summary

Technical Problem

Existing methods for predicting the risk of distant metastasis in nasopharyngeal carcinoma cannot fully utilize multimodal MRI image information, cannot effectively model the deep nonlinear interaction between clinical variables and imaging features, and are difficult to perform risk ranking learning on censored survival data, resulting in insufficient prediction accuracy and early identification of high-risk patients.

Method used

We employ a method based on cross-modal Transformer and depth consistency loss to construct a nasopharyngeal carcinoma distant metastasis risk prediction model by fusing multimodal MRI images and structured clinical data, utilizing cross-modal cross-attention mechanism for feature fusion, and combining depth consistency loss for risk ranking optimization.

Benefits of technology

It improved the accuracy and stability of predicting the risk of distant metastasis in nasopharyngeal carcinoma, enhanced the ability to identify high-risk patients, especially those with early distant metastasis, and improved the model's ranking ability in survival analysis.

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

Abstract

The invention relates to a nasopharyngeal carcinoma distant metastasis risk prediction method and system based on cross-modal Transform and depth consistency loss. The method comprises the following steps: acquiring multi-modal MRI image data and structured clinical data of a patient; obtaining a tumor area mask by using the pre-training model, cutting T1, T1C and T2 three-mode MRI images according to the tumor area mask, and constructing a region-of-interest image containing a tumor periphery microenvironment; inputting the image into a three-dimensional image encoder to extract deep image features, and performing embedded encoding on clinical variables; performing fusion modeling on the image and the clinical features through a cross attention mechanism to obtain patient-level comprehensive feature representation; and outputting a continuous distance transfer risk score based on the fusion feature, and performing joint optimization in combination with the auxiliary classification branch and the depth consistency loss to obtain a prediction result. According to the method, tumor and tumor-periphery microenvironment information is effectively focused, the multi-modal feature deep interaction and early high-risk identification capability is enhanced, and the accuracy of nasopharyngeal carcinoma distant metastasis risk prediction is improved.
Need to check novelty before this filing date? Find Prior Art