Lung cancer is the leading cause of cancer-related deaths in the United States. Both primary and metastatic lung cancer most commonly manifest as pulmonary nodules, which are readily visualized radiographically. CT scanning is currently the most sensitive non-invasive means available for detecting pulmonary nodules, but has suffered from limited sensitivity and high interobserver variability, particularly of smaller nodules. The recent development of multi-detector-row CT (MDCT) allows imaging of the lungs with unprecedented three-dimensional spatial resolution, up to 10 times greater than single-row CT systems within a single less than 10 second breathhold. For radiologists to harness the higher spatial resolution of MDCT data to improve lung nodule detection, they must overcome two key challenges - (1) time efficient interpretation of the 300-600 images that result from high-resolution MDCT scans of the lungs and (2) improve nodule detection sensitivity without losing specificity when examining 1-mm thick CT sections, where lung nodule and blood vessel discrimination is more difficult due to the greater similarity of their appearance when compared to thick-section acquisitions. The focus of this proposal, therefore, is to develop an optimized approach toward the detection of lung cancer with CT.
Our specific aims are: 1. To develop an automatic technique for detecting pulmonary nodules from lung CT data. 2. To determine the improvement in radiologist sensitivity and interobserver agreement for the detection of pulmonary nodules in patients suspected of having them when computer-aided detection (CAD) results are considered following initial radiologist assessment of CT images. We will optimize our CAD system by training on CT scans of pulmonary nodules obtained from two medical centers in different regions of the United States, and we will show that CAD can be as effective as a second radiologist in improving a radiologist's ability to detect pulmonary nodules on CT scans without substantially increasing falsely positive detections. Upon completion of this work, we will have enabled radiologists to take better advantage of the improved data available from MDCT scanners and substantially improve their ability to detect pulmonary nodules on CT scans, and thereby contribute to improvements in the management and care of patients with lung cancer.
Showing the most recent 10 out of 12 publications
