Automated detection of pulmonary nodules from whole lung CT scans

Lung cancer is the leading cause of cancer-related death accounting for 29% of all cancer deaths in the United States. Despite improvements in lung cancer treatment over the last several decades, about 95% of people diagnosed with lung cancer died within five years due to late diagnosis. It is well known that treatment of early stage lung cancer results in a substantially higher overall cure rate. The Early Lung Cancer Action Project (ELCAP) showed that in 80% of lung cancer patients, diagnosis could have been made much earlier. In the lung cancer screening process, radiologists analyze whole lung CT images of asymptomatic patients searching for nodules. CT scanners produce many thin slice axial images per patient. Hence, radiologists are confronted with the overwhelming task of interpreting a massive quantity of images. This has necessitated the development of an automated system.; The objective of this dissertation research was to develop and evaluate a robust algorithm that automatically detects different types of pulmonary nodules from whole lung helical CT scans. Three databases with whole lung CT scans were created to train and test the detection algorithm. Database A and Database B contained 250 sequentially selected scans with at least one solid nodule with 2.5mm and 1.25mm slice thicknesses respectively. Database C consisted of 100 1.25mm slice thickness whole lung CT scans with at least one non-solid nodule.; The solid nodule detection algorithm was trained and tested on Database A and B. For isolated nodules with sizes 4mm or larger, the algorithm achieved 94.0% sensitivity and 7.1 false positives per case (FPPC) for Database A (2.5mm). Similarly, the algorithm achieved 91% sensitivity and 6.9 FPPC for Database B (1.25mm). The algorithm achieved 92% sensitivity with 17.4 FPPC and 89% sensitivity with 5.5 FFPC for attached nodules with sizes 3mm or larger in Databases A (2.5mm) and B (1.25mm) respectively. The sub-solid nodule detection algorithm was trained and tested on Database C targeting large size sub-solid nodules. The algorithm achieved 87% sensitivity with 22.02 FPPC for sub-solid nodules ≥8 mm. In summary, the developed algorithm achieved practical performance for automated detection of pulmonary nodules. The automated system will be a useful tool to assist radiologists in identifying nodules from whole lung CT scans in a clinical setting.