Medical image recognition using artificial neural networks

Abstract EN

In this paper, an integrated artificial neural system has been designed to recognize and classify images.

The system consists of the following three stages: (1) converting images to vectors; (2) feature extraction, i.e., entering these vectors into Self Organizing Feature Map (SOM) network and (3) classification for which Learning Vector Quantization (LVQ) network is used.

The system has been applied to four classes of gray scale medical images to recognize and classify them.

A raw image is entered into the designed system without any preprocessing.

SOM network reduces image s dimensions and causes the information not to be influenced by the changes occurred in the image.

LVQ network has been compared with conventional methods (Minimum Distance Classification (MD) and K-Means Clustering Algorithm) considering recognition rate and recognition time.

LVQ network gives the best results (highest recognition rate about 94% and least recognition time depends on personal computer in use).