Computer Assisted Diagnosis System for CT Screening of Lung Cancer

点击翻译按钮获取中文摘要

In Japan more than 45,000 people died of lung cancer in 1997 and the lung cancer was the highest cause among males. Recently, several studies have been reporting feasibility and efficacy of helical CT imaging that it may be an effective screening modality for lung cancer. In order to improve the chance of detecting pulmonary nodules which may be early lung cancers, we have been developping a computer-assisted diagnostic (CAD) system. This CAD system has an automatic detection of candidate nodules based on low dose helical CT images. Double reading of the CT screening is one potential solution to decreasing the number of the missed nodules. However, in practice, that is not feasible. Another potential solution is the use of the CAD system as the second reader. Two thousand consecutive CT images, which are obtained as part of lung cancer screeing, were tested to the thoracic radiologists and the CAD system. As a result, many candidate nodules were detected by the CAD system which were not seen initially by the radiologist. The double reading between the radiologist and the CAD system was effectively certified. The CAD system also has a benign or malignant diagnosis for each nodule candidate based on high dose helical CT images. A hundred of consecutive CT images, which are measured for the detection of candidate nodules on the CT screening, were comparatively evaluated between the radiologists and the CAD system. The CAD system showed an 80 % performance superior to the radiologist interpretation. The CAD software may detect pulmonary nodules which may have otherwise been overlooked by the radiologists at the initial interpretation. This may provide substantial benefit to the patient if the detected nodules are found to be at an early stage of lung cancer.