Method for recognizing benign and malignant pulmonary nodules based on deep learning

Abstract

The invention relates to a method for recognizing benign and malignant pulmonary nodules, in particular to a method for recognizing benign and malignant pulmonary nodules based on deep learning. The method comprises the following steps: (1) preprocessing original pulmonary nodule electronic medical record data; (2) screening and classifying documents; (3) constructing text representation; (4) training a deep learning model; (5) adding an attention mechanism; (6) selecting a classifier to identify benign and malignant pulmonary nodules; (7) fusing model classification results. According to theinvention, the benign and malignant pulmonary nodules are judged by using text information to assist medical treatment; the related knowledge of deep learning is used for text classification, so the manual intervention is reduced, and the efficiency is higher; through different text feature input, the influence conditions of three factors including gender and age, current medical history and personal history on benign and malignant pre-judgment of pulmonary nodules are compared, and the final accuracy and stability of a classification and recognition model are improved by adopting a result fusion method.

Description

technical field [0001] The present invention relates to a method for identifying benign and malignant pulmonary nodules, more specifically, to a method for identifying benign and malignant pulmonary nodules based on deep learning. Background technique [0002] Solitary pulmonary nodule (Solitary Pulmonary Nodule, SPN) refers to a single, round nodule in the lung parenchyma, with a maximum radius of no more than 30 mm, without lymphadenopathy, atelectasis, pneumonia and other lesions. The benign and malignant judgment of solitary pulmonary nodules is very important, because the survival rate of early malignant pulmonary nodules is higher, but about half of the nodules that are surgically resected for unclear diagnosis are benign, which will cause Some unexpected serious consequences, for example, older patients will cause great harm to the body after the operation and even cannot withstand the operation process. Therefore, it is very important to predict the benign and malign...

AI Technical Summary

Problems solved by technology

The above-mentioned text classification based on electronic medical records mostly uses text information such as admission records, surgical records, and pathological reports in electronic medical records. The biggest difficulty is insufficient corpus, low generalization performance, and many manual interventions.

Images