DEVELOPMENT OF A CONTENT-BASED MAMMOGRAPHIC IMAGE RETRIEVAL SYSTEM

Abstract

Content Based Image Retrieval (CBIR) system is a significant approach that helps in the retrieval of image data from a large collection of images. The major goal of CBIR system in medicine is to retrieve images of similar features that lead to similar diagnosis. This solves the problem of variability in the interpretation of medical images that could lead to misdiagnosis. This research developed a content-based mammographic image retrieval system, with a view to providing radiologists with an efficient means for storing and retrieving medical images, as well as comparing current cases with previously diagnosed ones in a medical archive. The system developed employed a total of 1000 images, three hundred and twenty two images which were acquired from the mini-MIAS database and six hundred and seventy eight images from Digital Database for Screening Mammography (DDSM). The Gray Level Co-occurrence Matrix (GLCM) and Tamura feature extraction method were used to extract textural features from the images. Shape feature was extracted from the mammographic images using Edge Histogram Descriptor (EHD). Relevant features were selected from the extracted features using K-means algorithm. The features were optimized with Genetic Algorithm (GA). The textural similarity between the queried image and the images in the databases were computed using Euclidean distance measure, while the similarity between the shape features of the queried image and images in the database was obtained using edge method. The study adopts the Feature-Adaptive Relevance Feedback (FA-RF) mechanism to achieve an efficient image retrieval process. The model was implemented using Matrix Laboratory (Matlab). The performance of the system was evaluated using precision. The system gave a precision of 56.21% before optimization for DDSM and 82.84% after optimization. A precision of 63.85% was obtained before optimization for mini-MIAS database and 87.61% after optimization. The developed system was also compared with two existing CBIR systems. The precision obtained from the developed system and the existing one when used with the images in the mini-MIAS are 87.61% and 73.15% respectively. Also, the developed system and the existing system gave precision values of 82.84% and 75% respectively when used with the images in the DDSM. Based on the comparative analysis, the accuracy of the developed system is higher than the existing systems. The system will enhance evidence-based medicine by assisting radiologists to make timely and accurate decisions on results of similar pathologies.