A Novel CBIR Approach to Differential Diagnosis of Liver Tumor on Computed Tomography Images

Liver tumor is one of the deadliest diseases which can be cured without much difficulty if diagnosed in early stages. Content based image retrieval (CBIR) constitutes an important tool in computer aided diagnosis (CAD) which improves the diagnostic decisions of the radiologist by retrieving similar pathology bearing images from the medical database. In order to assist the radiologist in diagnosis of liver cancer, this paper proposes a novel CBIR based approach for differential diagnosis of liver tumor on computed tomography (CT) images as benign or malign. First, tumor is characterized by extracting its shape features using Fourier descriptors and texture features using MPEG-7 Gabor filter and edge histogram descriptors. Next, the dimensionality of the feature vector is reduced by applying principal component analysis (PCA).Finally, similarity matching process is accelerated using cluster-based indexing. The proposed approach was tested on medical image database consisting of 764 CT images of liver tumor. The experimental results demonstrate that the proposed method can effectively and efficiently retrieve similar case images from the database in response to query image.