| Research Abstract |
In this study, we developed some GMDH (Group Method of Data Handling)-type neural network algorithms which can automatically organize the optimum neural network architectures fitting the complexity of the various medical images such as MRI images, X-ray CT images, digital mammography, echo images and digital X-ray images and we applied these algorithms to the computer aided diagnosis (CAD). The GMDH-type neural network algorithms developed by us have an ability of self-organizing the optimum neural network architectures using the various neuron architectures such as sigmoid function type neuron, radial basis function type neuron and polynomial type neuron so as to fit the complexity of the various medical images. Furthermore, these algorithms have another ability of self-selecting the optimum input variables from many image characteristics so as to minimize the prediction error criterions defined as AIC (Akaike's Information Criterion) and PSS (Prediction Sum of Squares). Therefore, we can apply these GMDH-type neural network algorithms to the computer aided diagnosis (CAD) and the medical image recognition very easily.
In this study, we applied these GMDH-type neural network algorithms to the various medical images such as the MIRI image of the brain, X-ray image of the stomach and X-ray CT image of the lungs and we organized the optimum neural network architectures fitting the complexity of these medical images m the computer. By using these organized neural networks in the computer, the outlines of the interested regions (ROI) of these images ware automatically extracted with the good accuracy. Furthermore, these GMDH-type neural network algorithms are applied to the computer aided diagnosis of breast cancer.
|
