Construction of Detection Systems for Lung Cancer in Chest X-ray Images and Their Superior Performance

Abstract

We construct detection systems for lung cancer in chest X-ray images with neural networks. The exact position of abnormal area in the X-ray images, namely lung cancer is determined with the result from CT images, and then, the one-dimensional numeric sequences including the abnormal area are used as a teacher signal. In order to reduce noise included in the images, the two-dimensional median filters with various sizes are used in the pre-processing unit of this system, and their sizes are optimized with a view of the performance of the system. At the same time, the ensemble learning method is applied to the detection system. As a result, we were able to construct the detection system with high performance for lung cancer in chest X-ray images.