Automated methods for the analysis of nuclear medicine images could provide an objective diagnosis, and means to transfer sophisticated expertise to less experienced centres. The goal of this study was to develop software methods for the automated analysis of (a) Quality Control (QC) images, and (b) myocardial perfusion tomography images.;The system for the automated analysis of QC images was based on feature extraction algorithms, which provided input to a higher level diagnostic expert system. Several features characterizing QC images were defined. Rule-based and object-oriented expert systems were created to guide personnel in QC procedures, detect gamma camera faults, and suggest corrective actions. An object-oriented representation of knowledge allowed a natural representation and classification of image features, artefacts, and other concepts used in this knowledge domain. The feature extraction algorithms combined with a prototype expert system could perform diagnosis of gamma camera faults and QC procedure errors on a limited set of examples.;Computer-aided analysis of myocardial perfusion images was accomplished by creating three-dimensional (3-D) reference templates, to which patient's images could be automatically aligned using image registration algorithms. The templates included a normal distribution of activity and perfusion maps corresponding to specific coronary arteries. The quantification was done by a 3-D region-growing procedure that outlined perfusion defects in test-patients based on differences from the normal templates. Alignment and quantification methods of myocardial perfusion images were successfully tested on a group of 168 angiographically correlated patients. Perfusion defects were characterized in terms of numeric parameters, thus avoiding subjective visual assessment. The location of defects relative to the expected hypoperfusion sites was also established.;Analytical and artificial intelligence software methods can be used for automated interpretation of QC and cardiac images. Object-oriented methods are suitable for encoding the knowledge required for computer-aided analysis of QC images. A comprehensive and fully automated analysis of cardiac perfusion images is possible by comparison of patient data to 3-D reference models.
展开▼
