Abnormality detection for infection and fluid cases in chest radiograph

Abstract

This paper presents an automated abnormality detection system for infection and fluid cases in the lung field for chest radiograph. The abnormality features represented as abnormality scores are investigated based on the sharpness of costophrenic angle (Scoreθn), symmetrical lung area (ScoreLp), area of the lung (Scorearea), as well as the lung level (ScoreLlevel). The radiograph will be detected as abnormal if any of the score is `1'. Total numbers of classified normal and with disease radiographs are 177 and 35 respectively. From the results at the image level, 78% and 100% of the infection and fluid images are correctly detected as abnormal.