Abstract

This paper presents a quantitative evaluation of state-of-the art intensity based image registration with retrieval methods applied to medical images. The purpose of this study is to access the stability of these methods for medical image analysis. The accuracy of this medical image retrieval with affine based registration and without registration is evaluated using observer study. For retrieval without registration and with registration, we examine the performance of various transform methods for the retrieval of medical images by extracting the features. This helps for the early diagnosis. The technique used for retrieval of medical images were a set of 2-D  discrete Fourier transform (DFT), discrete cosine transform (DCT), discrete wavelet transform (DWT), Complex wavelet transform (CWT), and rotated complex wavelet filters (RCWF) were implemented and examined for MRI  imaging modalities. Especially RCWF gives texture information strongly oriented in six different directions (45° apart from the complex wavelet transform). Experimental results indicate that the DWT method perform well in retrieval of medical images. The method also retains the comparable levels of computational complexity. Then the experimental evaluation is carried by calculating the precision and recall values. It is found that DWT performs well for retrieval without registration and CWT with affine performs well in registration based retrieval with efficiency of 92% from retrieval efficiency 83% of DWT without registration. This helps in classification as before registration and after registration especially for clinical treatment and diagnosis.

Keywords

Registration, Retrieval, Content based image retrieval, Texture based medical image retrieval, Transforms