Head medical image auxiliary interpretation report generation method based on neural network

Abstract

The invention relates to a head medical image auxiliary interpretation structured report generation method based on a neural network. According to the method, the interpretation of the brain image andthe generation of the structured report can be completed in a full-automatic manner; the workload of a user is greatly reduced; the focus can be modified through manual intervention so as to improvethe accuracy; wherein the interpretation process is completed by stages; according to the method, the quality of each stage is guaranteed to be controllable. Meanwhile, corresponding interpretation models and structured reports are used for different image types and different disease types, compared with a report directly generated through images, the target is clearer, the result is more accurate, interpretability is higher, and the automatically generated report is more standard and objective and can be used for cultivating doctors with insufficient experience; the structured report is presented in a Key / Value form and stored in a JSON format, so that the report is convenient for doctors to modify, and the lightweight data format is convenient for storage and transmission.

Description

technical field [0001] The invention relates to medical image analysis in the field of computer image processing, in particular to a neural network-based method for generating a structured report for aided interpretation of head medical images. Background technique [0002] With the development of imaging technology, modern medical imaging such as CT (Computed Tomography) and MR (Magnetic Resonance) have been widely used in various hospitals, providing more and more support for disease screening and diagnosis. Radiologists need to review medical images and write image reports to describe the findings of the corresponding areas in the images. The writing of imaging reports requires an in-depth understanding of human anatomy, physiological characteristics, imaging analysis techniques, and clinical inference, and comprehensive consideration of various factors. Therefore, inexperienced doctors often need to go through lengthy training to be competent. At the same time, the writ...

AI Technical Summary

Problems solved by technology

[0005] First of all, existing methods lack detailed and accurate descriptions of lesions, and CNN's recognition of some global features is not ideal

Secondly, the presentation method of the report is very simple. The existing methods are not adaptable to images of different modalities and different parts. Usually, a model needs to be trained separately for a specific modality, part, and disease type.

Images