An investigation into the use of neural networks for an expert system in nuclear medicine image analysis

Describes exploratory work on the use of neural networks in building an expert system that will emulate the diagnostic skill of an experienced clinician in the area of nuclear medicine. The long term aim of the project is to produce an automatic reporting system which scans medical images, extracts relevant features, and gives a diagnostic output. Nuclear images contain a substantial amount of `noise´ and are difficult to interpret. Many of the imaging techniques give equivocal results, and diagnostic accuracy is not certain even in skilled hands. There is a lack of absolute rules which may be applied to determine whether an abnormality is present or not. The uncertainty in the decisions of image abnormalities and the incompleteness of data suggest neural networks, which are competent in areas of imperfect knowledge. A neural network will have additional advantages of speed, due to the inherently parallel nature of operation, and adaptability