Determining and classifying the region of interest in ultrasonic images of the breast using neural networks

This paper describes how ultrasonic images of the female breast have been processed and neural nets used to aid the identification of malignant and benign areas in them. The images are windowed, filtered and pre-processed into suitable patterns for processing by a neural net. Two networks are trained and used: one for malignant cases and the other for benign cases. These are used to make predictions of regions of interest which are presented as circles overlaid on the image. The system has been prototyped and tested and experts agreed well with the classification and localisation. The system is usually weak when the evidence on the image is considered weak by the expert. It is concluded that the system is promising and should be developed further by providing more training to the network.