Localization of the Suspected Abnormal Region in Chest Radiograph Images

Abstract

Chest X-rays (CXR) are widely used radio-imaging modality for preliminary diagnosis of thoracic abnormalities. Automatic detection and localization of diseases will greatly enhance real-world diagnosis processes. The imprecise and subtle appearance of disease responses on chest radiographs makes it difficult to localize sometimes even for an expert radiologists. The radiographic pattern in the normal regions is different from the disease affected regions. The affected areas in the X-ray images exhibit a cloudy pattern that shows the symptoms of various diseases like tuberculosis, pneumonia, pleural effusion, etc. The clustering algorithms have the capability to form clusters based on extracted features from the image. In this paper, we have investigated the effectiveness of Fuzzy C-Means (FCM) and K-Means (KM) clustering techniques to localize the suspected abnormal regions in CXR images. The analysis is performed on small image patches extracted from the segmented lung fields, and the results are compared with ground truth data provided by the radiologist. The proposed system is validated using a publicly available Montgomery dataset. The results demonstrate the promising performance of the proposed technique in delimiting the suspected abnormal regions.