Detection of pneumoconiosis rounded opacities using neural network

The fast detection of pneumoconiosis-affected parts using neural network is presented. The rounded opacities on the pneumoconiosis X-ray photo are picked up quickly through a backpropagation (BP) neural network with several typical training patterns. Training patterns from 0.6 mmØ to 4.0 mmØ are made as simple circles. The neck problem for automatic pneumoconiosis diagnosis is to reject unnecessary parts like ribs and blood vessel shadows. In this paper, such unnecessary parts are rejected well by a special technique called "moving normalization". The input of the neural network is a 30×30 pixel image which is quarried successively from a bi-level ROI (region of interest) image with a size of 500×500 pixels. The moving normalization technique has been developed in order to make an appropriate bi-level ROI image. A complete evaluation is carried out using a size and figure categorization. Many simulation examples show that the proposed method gives much more reliable results than traditional methods do