Madzin, Hizmawati (2013) Information fusion of text and visual features for multi-modality medical information retrieval / Hizmawati Madzin. PhD thesis, University of Malaya.
Abstract
The increase of multimedia data in medical field has led to the challenging problem of developing techniques that can provide efficient and accurate search and navigation through the large digital archives. Multi-modality of medical images such as x-ray, CT scan and MRI constitutes an important source of anatomical and functional information to provide valuable teaching and research, effective training and diagnosis of diseases. The approach of conventional text-based queries and exact matching with database is becoming obsolete. The mismatch of medical term between user query and document has become an issue. This drawback of conventional text-based retrieval motivates researchers towards more effective text-based retrieval and visual content-based image retrieval (CBIR) in medical field which has been an active research area in computer vision for the past few years. However CBIR solely has not yet succeeded in bridging the semantic gap between human concepts and low-level visual features. Information fusion support for human or automated analysis and processing relies on the hypothesis that the combination of multiple information sources allows for better results in information retrieval performance. This research is focused on information fusion of text and visual content analysis in medical information retrieval system. Initially the design, development and evaluation process are executed separately for text and content-based features. Medical hierarchical conceptual model is applied in both text and content-based frameworks which emphasize on modality, anatomy and pathology concepts. Multi-modality Medical Information Retrieval System (M3IRS) text-based framework consists of four main components which are document pre-processor, query processor, retrieval process and ranking strategies. Two ranking models are introduced namely Comprehensive and MedHieCon ranking models. Multi-modality Medical Image Classification System (M3ICS) is the content-based framework which is based on extracting visual features of texture, shape and color in global and local descriptors and applying semantic classification using MedHieCon model. Supervised learning technique is heavily used in visual classification to evaluate the performance of the framework. The final stage is the information fusion with the combination of text and content-based information sources. Hierarchical processing in late fusion technique is applied where the output from text-based processing will be the input for content-based system. Dataset from ImageCLEF 2010 medical task is used which includes 77,500 multi-modality medical images and documents. The performance of text-based, content-based and information fusion of text and content-based processing are evaluated. The effectiveness of text-based performance is better than content-based. However content-based approach is complement to text-based retrieval in order to increase the relevant documents into higher position ranking in medical domain information retrieval. Retrieval based on information fusion achieves the best performance by improving the effectiveness of retrieving relevant documents in medical domain.
Actions (For repository staff only : Login required)
![[img]](http://studentsrepo.um.edu.my/style/images/fileicons/application_pdf.png)
