Template matching-based ultrasound report generation methods, electronic devices, and software products

By constructing a fragment template library and using a bundle search algorithm to call template fragments one by one to generate ultrasound reports, the problem of template vocabulary dilution loss function in the existing technology is solved, which improves the accuracy of ultrasound reports, especially the description of pathological information.

CN122135877APending Publication Date: 2026-06-02SHENZHEN INST OF ADVANCED TECH CHINESE ACAD OF SCI
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SHENZHEN INST OF ADVANCED TECH CHINESE ACAD OF SCI
Filing Date
2026-03-24
Publication Date
2026-06-02

AI Technical Summary

Technical Problem

Existing ultrasound report generation technologies suffer from highly templated features, resulting in a large number of template terms that can be predicted without relying on ultrasound image information. This dilutes the loss function, makes it difficult for the model to focus on key pathological terms, and reduces the accuracy of report generation.

Method used

By constructing a fragment template library and using a bundle search algorithm to call template fragments one by one to generate ultrasound reports, the dilution effect of template vocabulary on the loss function is reduced, allowing the model to focus on key pathological terms.

Benefits of technology

It significantly improves the accuracy of ultrasound report generation, ensuring that the model can focus on key pathological information such as lesion location and echo characteristics.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122135877A_ABST
    Figure CN122135877A_ABST
Patent Text Reader

Abstract

This application relates to the field of machine learning technology, proposing a template-matching-based ultrasound report generation method, electronic device, and computer program product. The method includes: acquiring a target ultrasound image to be processed; inputting the target ultrasound image and a pre-built fragment template library into a trained ultrasound report generation model, and outputting a target ultrasound report corresponding to the target ultrasound image through the ultrasound report generation model; wherein, the fragment template library includes multiple template fragments obtained by mining ultrasound report corpora, each template fragment being a word, phrase, or clause; the ultrasound report generation model, based on a beam search algorithm, calls the template fragments one by one according to the matching results between the image features of the target ultrasound image and the text features of each template fragment to generate the target ultrasound report. This method can significantly reduce the dilution effect of template vocabulary on the loss function, enabling the model to focus on key pathological terms, thereby improving the accuracy of ultrasound report generation.
Need to check novelty before this filing date? Find Prior Art