Lung X-Ray Image Analysis for Automated Detection of Early Cancer and Tuberculosis

摘要

This paper reports our continuous work on the detection of lung cancer and tuberculosis for a computer-aided diagnosis (CAD) system. Preliminary diagnoses for the lung diseases are mostly based on X-ray images. This is a time consuming process. In some cases, medical experts had overlooked the diseases in their first examinations on X-ray images, and when the images were re-examined the disease signs could be detected. A CAD system can mark suspected areas on images for careful examination by medical doctors. Besides, early detection of the diseases is very important for successful treatment. Our current work aims at finding nodules, early symptoms of the diseases, appearing in lungs. This analysis is then used to sort patients in the order of their disease severities to help medical professionals to decide on examination schedules. We first used a modified Watershed segmentation approach to isolate a lung in an X-ray image, and then applied a small scanning window to check whether any pixel was part of a disease nodule. The preliminary experimental result showed that at least 50% nodules had been correctly detected with at most 25% false negatives.