Multi-modal nuclear magnetic resonance image case report automatic generation method

Abstract

The invention belongs to the medical data analysis and intelligent processing technical field, and specifically relates to a multi-modal nuclear magnetic resonance image case report automatic generation method employing a deep learning model; the method comprises the following steps: importing an attention matrix on the basis of using a convolution nerve network to extract image characteristics; giving different weights on characteristics of different positions via point multiplication operation, thus obtaining image characteristics under different attentions; using a long-short period memorycycle nerve network to form a subject vector of each sentence in a case report according to the image characteristics under different attentions; using another long-short period memory cycle nerve network to form each word according to the subject vector of the sentence; connecting the words so as to obtain the final case report. The method can automatically form a description text of a magnetic resonance image case without using a case template, thus providing deep meanings for mitigating works of radiology department doctors and building an intelligent computer auxiliary diagnosis system.

Description

technical field [0001] The invention belongs to the technical field of medical data analysis and intelligent processing, and in particular relates to a method for automatically generating medical record reports in natural language for nuclear magnetic resonance images. Background technique [0002] According to Hao Jie, President of the Cancer Hospital of the Chinese Academy of Medical Sciences, Director of the National Cancer Center, and Academician of the Chinese Academy of Sciences, at the annual meeting of radiation oncology on "The Status and Trends of Cancer in China", it can be found that the current cancer mortality rate in my country is It is higher than the global average of 17%. The reasons include the lack of an effective tumor screening mechanism and the lack of an efficient and homogeneous tumor diagnosis and treatment system, which hinders early diagnosis and effective later treatment of patients. Precision medicine and medical image-aided diagnosis systems bas...

AI Technical Summary

Problems solved by technology

In the traditional method, the classification of medical record attributes is obtained through image feature analysis. It is necessary to use a separate model for each attribute, and the results obtained are not readable text and lack of correlation ranking.

Images