A Content-Based Image Retrieval System for Osteo-Articular Applications

In this paper, we present a content-based indexing and retrieval (CBIR) system for osteo-articular MRI application, devoted to student formation and diagnosis help. A novel approach for image retrieval, called EMiner is introduced. It is based on a semi-supervised learning method using mixture models. We have also introduced a variable selection mechanism in the relevant feedback loop of our CBIR system. The performance of our indexing and retrieval method is assessed on a database of 395 images, extracted from different patient archives and labeled by an expert. According to the physicist supervision, the retrieval process is evaluated using a cross validation protocol and two quality criteria: precision and recall. Obtained results are satisfying. The proposed system is now implemented in the osteo-articular radiology service of Lille CHRU hospital