An essential factor in lung cancer survival is early detection, with up to 50% long term survival among patients whose tumors are detected at clinical stage I. Unfortunately, only 15% of lung cancers are currently detected at this stage. A major problem in early detection is the limitation of chest radiography, the initial procedure for detection of pulmonary nodules; it is estimated that 30% of positive nodules are currently missed. Therefore, the applicants proposed to develop an advanced computer-aided diagnosis (CAD) scheme that assists radiologists in detecting pulmonary nodules in chest radiographs by providing them with a """"""""second opinion."""""""" The main obstacle to achieving a clinically acceptable level of performance in such schemes is the large number of false-positive detections. Thus, this proposal focuses on improving the performance of pulmonary nodule detection by substantially reducing the number of false positives detected.
The specific aims are: (1) To establish image databases of pulmonary nodules for development and evaluation of a CAD scheme by (a)extracting regions of interest containing nodule candidates from an existing database, and (b) developing a new database of digital chest images for evaluation of the CAD schemes; (2) To develop methods for removal of normal anatomic structures by implementing radiologists' visual analysis strategy that (a) identify regions with similar normal anatomic structures in the lung, and (b) remove anatomic structures through non-linear registration and subtraction; (3) To develop methods to select and combine image features for reduction of false positives by (a) extracting image features characteristic of nodules from normal-structure-subtracted regions, (b) by applying adaptive rule-based tests, (c) combining features by artificial neural networks to obtain a single index for the reduction of false positives, and (d) using genetic algorithms to select features that are most effective in distinguishing nodules from false positives; and (4) To evaluate the benefit of the overall CAD scheme by (a) performing observer studies to estimate improvement of radiologists' diagnostic accuracy in the detection of pulmonary nodules without and with computer aid, and (b) performing pilot prospective studies for preclinical evaluation. The hypothesis is that a CAD scheme with a clinically acceptable sensitivity and specificity will substantially improve radiologists' accuracy in the detection of pulmonary nodules. Such a high-performance CAD scheme should advance the early detection of pulmonary nodules, and therefore has the potential to improve the prognosis for patients.
| Yoshida, Hiroyuki (2004) Local contralateral subtraction based on bilateral symmetry of lung for reduction of false positives in computerized detection of pulmonary nodules. IEEE Trans Biomed Eng 51:778-89 |
| Keserci, Bilgin; Yoshida, Hiroyuki (2002) Computerized detection of pulmonary nodules in chest radiographs based on morphological features and wavelet snake model. Med Image Anal 6:431-47 |
