A Classification Method for Mutation of Driver Gene Information from Thoracic CT Images Based on Machine Learning

Genetic testing confirms the mutation in driver gene information involved in cancer cell growth. If the mutation is identified, molecularly targeted drugs with significantly higher response rates and milder side effects are expected to play an active role, but due to the difficulty of identification by imaging findings, testing is performed by invasive biopsy. However, due to the difficulty of identification by image findings, the examination is performed by invasive biopsy. Therefore, CAD (Computer Aided Diagnosis) system is highly applicable to detect mutations non-invasively by applying the analysis results obtained from images. In recent years, the correlation between radiomics features and cancer has been confirmed, and the prediction, classification, and detection of lesions in unknown data by machine learning using these features have shown high performance. Therefore, with the goal of developing a non-invasive genetic testing CAD system, we propose a method for detecting driver gene information mutations from chest CT (Computed Tomography) images using Ensemble Learning. In this method, the radiomics features extracted from the chest CT images are used for supervised learning by ensemble learning. Then, the effectiveness of the proposed method is verified by classifying the images into different classes and performing evaluation experiments.