Image report context-aware intelligent completion method and system

Abstract

The application relates to medical data analysis and intelligent processing technology, and specifically discloses an image report context-aware intelligent completion method and system. First, historical image reports for various cases are acquired, a completion training data set is constructed, and a certain proportion of words, sentences or paragraphs are randomly masked. Then, medical knowledge is integrated into an open-source generative AI model through a continued pre-training technology, a full-parameter fine-tuning strategy is used for training and optimization, and a context-aware intelligent completion model is obtained. Finally, in the actual reasoning stage of the context-aware intelligent completion model, a TOP-K or adjusted temperature strategy is used to obtain several preferred completion schemes. The application realizes intelligent automatic completion of the current incomplete part at the word, sentence and paragraph levels based on the understanding of the report context by the model, and aims to effectively improve the work efficiency in the report review or fine-tuning stage, and reduce the time and energy costs of manual writing and revision.

Images