An Efficient CBIR System for Medical Images Using Neural Network

Abstract

This paper introduces an innovative Content-Based Image Retrieval (CBIR) system that has been specifically developed for medical databases. Its objective is to resolve the drawbacks of conventional keyword-based search approaches when considering the widespread digitization of medical illustrations, diagrams, and paintings. In contrast to conventional methods that rely on textual queries, CBIR systems effectively locate and retrieve relevant images by analyzing image content using computer vision and image processing techniques, as well as information retrieval and database methods.
A key challenge in CBIR lies in bridging the semantic gap between high-level user queries, often expressed through example images, and the low-level features of images such as texture, shape, and objects. This paper explores techniques to mitigate this disparity, enhancing the system's ability to accurately interpret user queries and retrieve relevant images.

The proposed CBIR system operates within a medical database containing images of various human organs, including the brain, heart, hand, chest, spine, and shoulder, categorized into six distinct classes. By leveraging low-level image features such as texture and shape, extracted using methods like mean, variance, standard deviation, area, perimeter, circularity, and aspect ratio analysis, the system performs iterative searches to retrieve relevant images.
Classification of retrieved images is accomplished using Artificial Neural Networks (ANN), which have demonstrated efficacy in medical image classification tasks based on imaging modalities and the presence of normal or abnormal conditions. Performance evaluation of the CBIR system is conducted using confusion matrices to calculate precision and recall, essential metrics for assessing retrieval accuracy.

By focusing on medical datasets and integrating advanced feature extraction and classification techniques, this CBIR system aims to significantly enhance image retrieval efficiency and accuracy, particularly in the context of medical applications where precise retrieval of relevant images is critical for diagnostic and research purposes.