Summarizing Medical Imaging Reports With Lstm Deep Learning Method for Effective Diagnosis / Treatment Process

Abstract

Abstract The study aims to summarize the medical imaging reports automatically and to use them effectively in diagnosis and treatment. Summary reports will save time and reduce the workload by helping specialist physicians. Since summary reports will be more likely to be read in full instead of lengthy reports, the possibility of missing important details will decrease. 2457 medical imaging reports from 2199 people obtained from Medical Imaging Centers were used as data in automatic summary (retrospective patient records from 2019–2022). In the study, a model based on natural language processing, deep learning, and sequence-to-sequence architecture was designed. The success of summaries obtained automatically from medical imaging reports was evaluated with the ROUGE metric. In the study, all of the reports and the summaries of the report by the doctors were examined. The summary of the doctors was taken as a reference and compared obtained summary. The critical findings in the reference summary created by the physicians are also included in the summary obtained in the study. With the continuous increase in data in the health field, the need for summary systems is also increasing. It aims to effectively use the summary reports obtained in this study in the diagnosis and treatment process. Thus, patients' past and current imaging results will be compared quickly, and all current and previous reports will be used effectively for more accurate diagnosis and treatment.