Computer-aided diagnosis in two-phase 201Tl-SPECT of thoracic lesions.

AIM: The retention index, a traditionally quantitative analysis of two-phase (201)Tl single photon emission computed tomography (SPECT) of the chest, is manually calculated by experienced physicians from comparable 2-D ROI. However, a 3-D ROI would provide more information than a 2-D ROI extracted from a single frame of SPECT. We propose a new diagnostic system, computer-aided diagnosis (CAD), to automatically detect suspicious lesions as 3-D objects on chest (201)Tl-SPECT, and assist the physician in interpreting these images. PATIENTS, METHODS: Seventy patients with thoracic lesions and confirmed diagnoses were enrolled to test this automatic CAD system. The reliability of the CAD system in detecting lesions as 3-D objects was compared to the 2-D ROI of (201)Tl-SPECT found by the manually visualized method. Furthermore, we also proposed a novel index, the retention index using the heart (RIH), to differentiate high retention (slow clearance, increasing target to heart ratio) as a criterion for a malignant lesion, from low retention (faster clearance, small or no increase of the target to heart ratio) for benign lesions. RESULTS: The CAD system can achieve a detection rate of 100% in automatically searching for thoracic lesions in (201)Tl-SPECT. In diagnostic performance, the CAD system with the RIH of comparable 3-D objects has an area under the ROC curve of 0.86, higher than the 0.78 of the traditional RI method (p=0.198). CONCLUSION: The CAD system of two-phase (201)Tl-SPECT is a promising tool for detecting and diagnosing thoracic lesions with a diagnostic accuracy of 0.81.