Computer Aided Diagnosis of Lung Cancer

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Lung cancer is the leading cause of cancer deaths in both men and women. A 70 percent five-year survival rate has been reported when the lung cancer is diagnosed at a local stage, compared to 2 percent when distant metastases are found. Recent studies indicate that helical CT may be an effective screening tool for lung cancer. The American College of Radiology Imaging Network (ACRIN) will begin a randomized controlled trial of helical CT for lung cancer screening to evaluate its efficacy. Analysis of CT images to detect lung nodules is a demanding task for radiologists. Some lung nodules will likely be overlooked because of the overwhelming amount of information to be interpreted. Characterization of detected nodules to reduce unnecessary biopsies will also become more important as the number of thoracic CT exams increases. Computer-aided diagnosis (CAD) can be a viable approach to improving the accuracy and efficiency of lung cancer detection in CT images. It will be particularly useful if lung cancer screening with CT is implemented. The goal of the proposed project is to develop a CAD system for early detection of lung cancers on thoracic helical CT images. We hypothesize that an accurate CAD system (1) can be developed, (2) can be used as a second opinion to assist radiologists in interpretation of thoracic CT exams, and (3) will improve radiologists' accuracy for lung cancer detection. We will develop advanced computer vision techniques to automatically segment helical CT images, detect candidate pulmonary nodules, differentiate nodule and normal pulmonary structures, and estimate the likelihood of malignancy of the nodules. Computerized image segmentation and feature extraction techniques will be developed based on expert knowledge and image characteristics. Statistical classifiers, fuzzy classifiers, and artificial neural networks will be designed to differentiate nodules and normal structures, as well as to characterize malignant and benign nodules. Quantitative CT phantom studies will be performed to develop reliable methods for estimating the calcium concentration of nodules and estimating nodule volume on CT images so that these features can be used in our CAD system for malignancy detection. Observer performance studies using receiver operating characteristic (ROC) methodology will be conducted to evaluate the effects of CAD on radiologists' detection and classification of lung nodules in CT images. A large public database of helical CT cases to be collected by an NIH-supported consortium will be the main data source for the development of the computer vision techniques. The innovations of this project include: (1) developing region-specific computer vision method for detection of lung nodules; (2) eliminating the vascular tree in the helium for false positive reduction, (3) exploring interval change analysis for classification of malignant and benign nodules; (4) developing a quantitative method for measuring the calcium concentration of nodules for improved characterization of calcified nodules, and (5) performing ROC studies to evaluate CAD's ability to assist radiologists in the detection and characterization of lung nodules in CT studies. It is expected that the proposed studies will result in an effective CAD system for lung cancer diagnosis. When fully developed and clinically implemented, a CAD system for lung nodules will increase the efficacy of lung cancer screening with helical CT, improve early detection, and improve the chance of survival of patients.