Boosting biomedical images indexing

Indexing and retrieval in biomedical image databases is a challenging problem. Constructing large-scale indexing solutions is typically limited by a choice of appropriate features, complexity constraints of the engine and a way how to combine retrieval results to have a stronger one. Combination of standard feature extraction routines with specific knowledge on a subject, such as precise automatic object segmentation and medical parameters estimation is the first key factor to achieve high accuracy and robustness of the indexing/retrieval solution. We are developing a search engine based on a TTA10 algorithm, which stores data in hierarchical fashion, with logarithmic complexity to access a large data repository in real-time. We propose to use AdaBoost technique to combine independent search results into more robust and accurate one. Initial results on a database of more than 80.000 ultrasound images demonstrate good accuracy and fast speed.