Research on Image Perception of Abnormal Findings

点击翻译按钮获取中文摘要

We have studied how radiographic noise and anatomic noise will affect the detectability of subtle abnormal findings.The influence of noise level on the detection performance of simulated lung nodules in the Gaussian white noise background has been investigated by using a four-alternative forced choice method. Results were mostly in agreement with the ones derived from the signal detection theory. The nodule detectability had little relation to the fundamental statistical measures of gray scale values of the local surrounding anatomic strucures, and was related to the signal-to-noise ratio.Next, we have accessed the influence of anatomic noise on the detectablity of subtle lung nodules depicted on chest radiographs, from a local aspect in which the existence of rib structure surrounding nodules is the focus. The method proposed by Samei et al. was applied to our study. From the normal chest radiography images, square regions were extracted by considering rib structure ; the centers of these square images were on the upper margin of a rib, the inside of a rib, the lower margin of a rib, and the center between two adjoining ribs ; and the simulated nodules were digitally superimposed on the centers of these extracted square images. By using a set of these processed images, we have performed the image reading study. Results indicated a significant effect of the existence of rib structure surrounding the nodule on the detection performance, and the detectability of the nodules on the center between two adjoining ribs was significantly the best. Further, the detectability of the nodules other than on the center between two adjoining ribs showed a tendency to decrease in the order of the upper margin, the inside, and the lower margin of a rib ; however, these differences were few. Radiographic noise of a certain size can also influence the nodule detection.