Lung cancer is a leading cause of death in the United States. The National Lung Screening Trial (NLST) showed that more lung cancers can be detected at an early stage with low dose CT screening. However, over-diagnosis of indolent lung cancer and benign nodules is one of the major limitations of screening, resulting in unnecessary treatment, biopsy, follow-up, increased radiation exposure, patient anxiety, and cost. Due to a lack of in-depth knowledge of the correlation of structural image features and histologic findings of lung nodules and the absence of validated diagnostic biomarkers for accurate disease categorization, the current diagnosis and management of the screen-detected nodules remains challenging. The goal of this proposed project is to develop a decision support system (DSS) based on quantitative histopathology correlated CT descriptor (q-PCD) of pulmonary nodules using advanced computer vision and machine learning techniques to characterize the histopathologic features of nodules and analyze their correlations with CT image features for improvement of early detection of lung cancer. We hypothesize that the proposed q-PCD analysis will have strong association with histopathologic characterization, and therefore will be a more effective biomarker for differentiation of invasive, pre-invasive, and benign nodules than conventional image-based features or radiologists' visual judgement. Accurate characterization of the nodule types will assist radiologists in making decision for management of the detected nodules; e.g., enabling early detection and treatment of invasive lung cancer, safe surveillance or replacing lobectomy with limited sublobar resection for pre-invasive lung cancer, and sparing biopsy of benign nodules, thereby reducing morbidity and costs in lung cancer screening programs. Our major specific aims are to 1) collect a large database of LDCT screening cases from NLST project and our institute to develop automated image analysis methods, 2) to develop a new DSS based on quantitative pathologic correlated CT descriptors (q-PCD) of lung nodules, 3) validate the effectiveness of DSS in lung cancer diagnosis. To achieve these aims, we will collect a large data set from the National Lung Screening Trial (NLST) and our institute. The collected database will include the baseline and follow up scans, pathology data, demographic information and other information provided by NLST. We will develop automated segmentation methods to extract the volumes of the solid and sub-solid components of detected lung nodules, develop quantitative methods to characterize the radiologic and pathologic features of lung nodules as well as the surrounding lung parenchyma, develop a novel radiopathomics strategy to correlate pathomics with radiomics, and to identify new imaging biomarkers. We will develop a clinically-translatable DSS with a joint biomarker combining both image and patient information, and evaluate its performance in lung cancer diagnosis, including its effectiveness in baseline screening CT exams and in follow up exams.
Lung cancer is the second most commonly diagnosed cancer in both men and women in the United States. If successful, we will have developed an automated low dose CT-based quantitative histopathology correlated CT descriptor (q-PCD) which can be included as an independent factor to improve the diagnosis of screening detected nodules. It will not only provide a useful imaging biomarker to assist in lung nodule management for individuals, but also constitute a strong foundation to meet our long-term goal of developing clinically- translatable decision support system (DSS) to facilitate personalized medicine in early detection of lung cancer, optimizing treatment strategies and reducing health care costs
