ROC analysis of 3D X-ray CT performance for lesion detection

Task based performance and image quality evaluation has been widely researched in recent years. Most of works focused on 2D nuclear medicine imaging. Since X-ray CT dosage has been pushed to lower level and 3D imaging has been a trend, it would be intuitive to study the task based performance for 3D X-ray CT systems. In this work, we investigate the application of ROC analysis of lesion detection from 3D X-ray CT images. Using the methodology of classical ROC analysis, we studied the difference in performance of detecting hot and cold lesions in X-ray CT. Moreover, to explore the advantage of 3D imaging, we proposed our strategies of combining information from orthogonal slices comparing with reading fully 3D visualization of volumetric images. We quantitatively evaluated the improvement of performance we can obtain from 3D data comparing with using 2D images only. Also, 3D visualization of volumetric images will greatly improve the detection performance.