Medical image modality classification using feature weighted clustering approach
Loading...
Date
2010
Authors
Cchandra, Bhavik Anil
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Medical Image Retrieval System is an area of great importance to the healthcare providers.
With a variety of modalities in use, it is important that retrieval systems can provide filtering
of results based on modality. The current method of identifying modalities is not accurate and
often wrong modality is assigned to the medical images. The objective of this work was to
propose a methodology to classify existing medical images by inspecting their visual features
and reducing human intervention. This thesis proposes the clustering of medical images using
MPEG 7 Visual Descriptor, SCAD clustering with simultaneous feature discrimination and
using k-Nearest Neighbour classification. Using this proposed work, the features that are relevant
to the cluster are assigned higher weights, giving importance to a number of features that
represent the clusters. The features that are not significant to the cluster are not totally ignored
instead given lower weights. Using this method, the medical images will be clustered and categorized
into their respective modality based on their visual features. In this thesis, emphasis
is given to Edge Histogram Descriptor, Color Layout Descriptor and Homogeneous Texture
Descriptor of MPEG-7. The proposed method is tested against Fuzzy C-Mean clustering. This
thesis also concludes that medical image modality classification is possible depending on the
strength of the feature descriptors to detect the characteristics of the medical images.