Shape-Based Medical Image Retrieval System

This article introduces my undergraduate thesis project focused on developing a shape-based medical image retrieval system. In this system, we implemented wavelet denoising techniques during the preprocessing stage to effectively reduce noise in medical images while preserving important shape characteristics. The core feature extraction methodology employs invariant moments (specifically Hu moments), which provide rotation, scale, and translation invariance - crucial properties for reliable medical image matching. We further enhanced the system by incorporating advanced algorithms and techniques, including shape matching algorithms that compare contour signatures using dynamic programming approaches, and deep learning-based feature extraction methods utilizing convolutional neural networks (CNNs) to capture hierarchical shape representations. The system architecture involves preprocessing modules, feature extraction pipelines, and similarity measurement components implemented through optimized matrix operations and distance calculations. Overall, this article summarizes the research and development work conducted during my graduation project, aiming to contribute to the field of medical image retrieval research and applications. Our implemented system not only assists medical professionals and researchers in faster, more accurate retrieval and diagnosis of medical images but also demonstrates practical utility and scalability for clinical environments. The code structure includes modular components for image processing, feature computation, and database management, ensuring maintainability and extensibility for future enhancements.