Application of neural networks to the processing of medical images

Abstract

Neural networks are applied to the enhancement of medical images. The purpose is to reduce computation time and achieve real-time capability. A tomography method is chosen, as most medical images are tomographic in nature. Two approaches are taken. The first approach starts with applying neural networks right from the beginning during the recording of the scattered wavefronts. The second approach is to apply neural networks only after the reconstruction of the image for image processing. The Hopfield model of neural networks is used for both approaches. The procedures for both cases are given. A detailed analysis for both cases is given, and a study is made concerning which approach will give the shorter computation time. Preliminary estimates show that when applying the neural network right from wavefield recording in the beginning, one needs about 10/sup 4/ iterations to resolve two object points. If the neural network is applied to process the image after the reconstruction, about 200 iterations are required for an L=256 and M=256 image. Hence, it takes less computing time for the latter approach. >