Classification and detection of lung nodules using virtual dual energy in CXR images

Major challenges in the detection nodules in chest radiographs (CXRs) are difficult because of the overlap with ribs and/or clavicles and the false positives (FPs) occurs because of ribs. Detection of nodules is very important, because radiologists will miss the subtle nodules. In this proposed system the ribs and/or clavicles are suppressed, so the false positives will be reduced. The ribs are suppressed by “virtual dual energy” (VDE Technology). Gray-level-based features were extracted from each candidate from both original and VDE CXRs. The CXRs images are collected form database JSRT (Japanese Society of Radiological Technology). Lung tumors at an early stage were detected by computed tomography (CT). It is a more sensitive than chest radiographs (CXRs). The 30% of nodules are missed because of ribs and/or clavicles in normal CXRs. By using Virtual Dual-Energy (VDE) Radiography the missing nodules can be detected. This improves the sensitivity, accuracy, specification and reduces incorrect prediction of abnormal conditions (false positives).