Classification of Histological Types of Primary Lung Cancer from CT Images Using Clinical Information

Identification of primary lung cancer is very important because it influences the course of treatment, especially for small cell carcinomas, which metastasize rapidly and must be detected at an early stage. In addition to imaging, clinical information is often used in CAD (computer aided diagnosis) systems. In addition to images, clinical information is often used in CAD systems, especially information on smoking history, which is considered to be important in the diagnosis of lung cancer. In this paper, we propose a method to identify primary lung cancer by adding clinical information from medical records in addition to images in order to improve the accuracy of diagnosis. We use tumor images surrounded by rectangular regions from CT images in an open dataset as input images and train the method by deep learning. We evaluate the proposed method by discriminating tumors from unknown data. In our experiments, we found that the accuracy was improved by about 5% when clinical information was added to 655 images, which included four classes of cancer: adenocarcinoma, small cell carcinoma, squamous cell carcinoma, and large cell carcinoma.