Colon cancer is the second leading cause of cancer deaths for men and women in the United States. However, it would be prevented by early detection and removal of its precursor lesions. The use of CT colonography (CTC) would substantially increase the access, capacity, safety, cost-effectiveness, and patient compliance of colorectal examinations. The interpretation of CTC examinations would be most effective by use of a first-reader computer- aided detection (FR-CADe) paradigm, where a radiologist reviews only the lesion candidates detected automatically by a computer-aided detection (CADe) system. However, because CADe systems can miss large masses, radiologists still need to perform an additional two-dimensional (2D) review of the CT images of the colon, which increases reading time over 40% on average. Furthermore, also radiologists can occasionally miss some types of masses on CTC images. The goal of this project is to develop a DEep RAdiomics LEarning (DERALE) scheme for the detection of large masses on CTC images. The scheme will be used to integrate deep learning methods and radiomic biomarkers to perform a complete automated review of CTC images for reliable detection of colorectal masses. We hypothesize that the DERALE scheme will be able to detect colorectal masses at a sensitivity comparable to that of unaided expert radiologists and that it can be used to reduce the interpretation time of FR- CADe without degrading diagnostic accuracy in CTC. We will evaluate and compare the classification performance of DERALE with that of unaided expert radiologists and conduct an observer performance study to compare the detection accuracy of the use of DERALE in the FR-CADe paradigm with that of unaided expert radiologists in the detection of masses from CTC images. Successful development and broad adoption of DERALE in the FR-CADe paradigm will facilitate early, accurate, and cost-effective diagnoses, and thus it will reduce the mortality rate from colon cancer, one of the largest threats of cancer deaths in the United States.
Successful development and validation of DERALE will substantially advance the clinical implementation of CTC and the FR-CADe paradigm in large populations to facilitate early, accurate, and cost-effective diagnoses, and thus it will reduce mortality from colon cancer, the second leading cause of cancer deaths in both men and women in the United States. The research is innovative in that robust mass detection has not been developed for CTC and FR- CADe, and that there have been no attempts to use deep learning for mass detection in CTC.