Low dose CT screening (LDCT) has been approved by the Centers for Medicare & Medicaid Services (CMS) for Medicare coverage. However, the primary problem associated with CT is the high false positive detections (~96%). This issue often leads to unpleasant and costly unintended consequences (e.g., follow-up scans and/or invasive biopsies). In this project, we propose to develop and commercialize a novel computer tool to aid in accurate assessment of indeterminate lung nodules. The goal is to accurately and efficiently quantify the potential risk of developing lung cancer and its future prognosis, thereby facilitating precise / personalized lung cancer screening and optimal patient management. During Phase I of this project, we have accomplished the proposed milestones and developed a prototype system that is now publicly accessible. Our preliminary validation of the prototype system demonstrates very promising performance. In Phase II, we will continue our effort to make this tool available to serve clinical community with the following specific aims: (1) develop a generalized framework that supports the incorporation of both image and patient information as well as other biological tests related to lung cancer; (2) fully optimize the computer algorithms to efficiently analyze chest CT scans and in particular synergize our algorithms with the deep learning technology to improve training efficiency and benign/malignance classification accuracy; and (3) comprehensively validate and optimize the system in clinical environment at the University of Pittsburgh Medical Center (UPMC). We believe that the proposed system is extremely timely and important in light of the CMS decision to cover annual LDCT lung cancer screening. Its availability will significantly reduce the over-diagnosis associated with LDCT for lung cancer screening and relieve the economic burden on healthcare system.
The project aims to develop an online computer tool to aid in efficacious early lung cancer screening and diagnosis based on LDCT. This tool will not only locate the pulmonary nodules automatically, but also quantitatively assess the malignancy of a suspicious nodule and its prognosis.