Analyzing DICOM and non-DICOM Features in Content-Based Medical Image Retrieval: A Multi-Layer Approach Antonio da Luz Jr., Daniel D. Abdala, Aldo v. Wangenheim, Eros Comunello Cyclops Project - Telemedicine Lab - Federal University of Santa Catarina {antoniol, caju, awangenh}@inf.ufsc.br, eros@telemedicina.ufsc.br Abstract This paper presents a hybrid approach to perform content-based retrieval on medical image databases. It takes advantage of a pre-processed case base that is batch updated. DICOM information is used to perform pre-filtering to speed up the retrieval process and a image processing knowledge base is used to dynamically reconfigure the most appropriated image processing procedures to perform the image feature extraction. It shows that pre-filtering can speed up considerably the retrieval process and also that some image features produce very similar results what leads to future work on defining the needed digital image processing knowledge base. 1. Introduction This paper presents new techniques to perform content-based image retrieval from DICOM-compliant medical image databases, especially new procedures to execute medical image retrieval. We work on images stored using the DICOM [1] – Digital Imaging and Communication in Medicine – standard. The restriction to image retrieval using only DICOM compliant images has as main reason, namely the usage of specific DICOM tags to perform a pre filtering and classification; focusing on more accurate results as well as to carry out the retrieval task on smaller image sets and shorter timeframes. The proposed methodology was idealized as a multi-layer architecture, intending to obtain quicker and more efficient image retrieval, filtering the images that must be analyzed in each step based on several different contextual information. Other proposals to solve the problem of context based medical image retrieval have been performed in the past [2, 3, 4, 5, 6 and 7]. These approaches can be classified in two different groups: (a) intended to work on a specific modality of images only, showing always a specific anatomical structure, e.g. CT-head or MR- knee, with very specific attributes. This technique is used in medical diagnosis applications and the identification and retrieval techniques used are application dependent; or (b) approaches concerned with more generic methods that process the images using generic image processing techniques. These approaches are normally applied on image retrieval from larger image databases and can or cannot be specific for medical applications. Our work intends to present a hybrid approach. As in (b), we work on very larger image databases, more specifically DICOM image databases, but the techniques necessary to perform the examination identification and image retrieval are examination modality dependent as those classified in (a). The first step of our approach is concerned with pre-processing the query image extracting all relevant information contained in DICOM tags and to raise the necessary search decision attributes. A pre-filtering procedure is executed with part of the retrieval parameters gained based on DICOM tag information, filtering the database to sort out the examinations that will not fit the query parameters. The next step is concerned in configure the image content identification parameters based on the same DICOM retrieval attributes used by the previous step. A novel approach is used to perform the query over the search space. The DICOM image database is batch pre-processed using the fitting techniques and an intermediary data structure – IMAGE – is created. The query over the search space is then executed over a database composed only by the preprocessed IMAGE structures. This strategy speeds up the search even more, but imposes a constraint of a previous pre- processing of the entire database. The results interpretation is matched with the remaining retrieval attributes, and finally, the results are presented with its matching percentage. 2. Material & Methods Proceedings of the 19th IEEE Symposium on Computer-Based Medical Systems (CBMS'06) 0-7695-2517-1/06 $20.00 © 2006 IEEE