Long Term Goal: Automatic screening of computerized tomography (CT) images of the head by (i) the use of wavelet frame image representations to identify candidate anatomical structures and (ii) use knowledge based system (KBS) constraints, under a Blackboard control to classify structures with a 3D model.
Hypothesis: The type of knowledge and matching strategies used can be more important than information content alone for recognition of anatomical structure in 3D data sets.
Specific Aim of This Pilot Study: To demonstrate that automatic recognition of large scale anatomical features in CT head images is more dependent on model construction and matching strategies than image content alone.
Health Relatedness: Automatic image recognition of CT images should improve diagnostic accuracy without significant cost increases. It would need a system which could classify patients as normal or abnormal which is the first major step in the diagnostic process. Abnormal images would be flagged prior to a radiologist's viewing. The radiologist, not the computer, would make the diagnosis. In essence the computer would be contributing to a double viewing method which has been shown (by radiologists) to improve diagnostic accuracy. This benefit should decrease health costs and morbidity.
Research Design and Methods: 30 sets of adult axial CT images have been subjectively examined to identify image features to describe the maxillofacial region of the head. An object-centered coordinate system (OCCS) is automatically generated using large scale reliable automatic features. General anatomical spatial relationships, which relate to the OCCS, have been manually entered into the KBS model. The image sets have been processed to form multiscale wavelet representations producing two subsets, one with high and the other low image information content images and different search strategies.
Testing KBS Structures: 30 patient data sets (containing high and low information wavelet representations) will be repeatedly examined with fixed set of KBS data, rules and matchilabels is identifying anatomical features which are either true or false. Anatomical feature recognition rates by alteration of KBS and matching strategies will be recorded. Also the effect on recognition rates of using identical high and low information content images will be observed.
This work is supported by N.I.D.R. Grant #1 RO3 DE10878