Lung Nodule Detection on Rib Eliminated Radiographs

摘要

A lung nodule detection algorithm was developed and tested against a comprehensive set of radiographs. Two new algorithms were utilized in the detection scheme. A preprocessing step eliminates ribs and collarbones on the image to enhance the visibility of nodules. The next step uses the Constrained Sliding Band Filter (CSBF) to raise the intensity of round shaped objects while suppressing other areas. The suspicious areas are then processed by a Support Vector Machine (SVM) based on mostly textural features to reduce the number of false detections. The algorithms were tested on the public database created by the Japanese Society of Radiological Technology (JSRT) and a private database. The new methods showed promising results, while the overall performance, 61% sensitivity at 2.5 false positives per image is comparable with state-of-the art algorithms.