Research On Medical Imaging Data Mining Based On Wavelet Transform And Neural Network

Worldwide, breast cancer is one of the diseases of high incidence and mortality rates, posing a serious threat to human health. Earlier diagnosis and treatment of breast cancer are critical to increase survival rates. Currently, the primary methods of diagnosis of breast cancer include X-ray images, CT, MRI and so on, and experts discern medical images with the naked eye. Due to lack of effective auxiliary technologies to improve the diagnosis accuracy, diagnostic omission errors are on high levels. Worse still, some early-stage breast cancer patients miss the best treatment. With the development of computer-aided medical technology, computers are widely used in secondary medical diagnosis.In this paper, key technologies and algorithms, such as wavelet, rough set and artificial neural network theory, which are applied to medical imaging, are studied, and wavelet neural network classification algorithm is optimized and applied to mammograms for breast cancer diagnosis . The main research includes:1. Improved wavelet neural network algorithm Improved Wavelet Neural Network classifier algorithm is proposed and designed in this paper. Take full advantage of the characteristics of wavelet transform, through the translation of scaling and multi-scale image signal analysis, can effectively extract the signals of local information; the neural network has self-learning, adaptive and fault tolerance features, is a kind of universal function approximator. Therefore, improved wavelet neural network has a stronger approach and tolerance. New mathematical method of data mining algorithms have high accuracy and distinguish error rate less. Improve the efficiency of image mining based on the improved results.2. The improved wavelet neural network classifiers in medical image Mining To improve medical imaging classifier performance, wavelet neural network classifiers, based on wavelet and artificial neural network theory, has been designed. The classifier has such advantages as good approximation ability, high convergence speed, etc. Better convergence rate and higher classification accuracy will be achieved when wavelet neural network is applied to MIAS dataset study, and its average error recognition rates are close to 90%.3. Wavelet neural network based on rough set In order to reduce the training time of neural network classifier, and further improve its classification performance in the meanwhile, wavelet neural network classifier algorithm based on rough sets is proposed and designed in this paper. It aims at exploring how to use the reduction principle of information gain based on rough set theory, and optimize feature attributes, thus it'll take less time to train the neural network and over reduction, which will happen when rough set classification is used separately, will simultaneously be avoided. After the proposed classifier is evaluated over MIAS datasets---mammographic datasets, and back propagation(BP) neural network classifier and neural network classifier are compared, wavelet neural network classifiers based on rough sets is found to have better classification performance, meanwhile, take much less Training time, thereby, improve diagnosis accuracy.