MRI Brain Abnormality Detection Using Fuzzy Neural Networks
Paper i proceeding, 1996
In this paper an expert system for detection of brain abnormalities is proposed. First
preceding methods for segmentation of MR images are reviewed and their limitations are
discussed. In the proposed method, MR images (three images from one slice: T1, T2 and
Proton Density) are acquired from a scanner or directly from MRI system. For noise
deletation two filters (median and bandreject lowpass) are used (This stage is optional).
They make a clean view of MR images. It is necessary to have precise detections. So by
implementing a gray-scale to color transformation algorithm (it is a radially symmetric
butterworth band-reject filter), system can recognize the differences between tissues
accurately. Now we have three colored images (T1, T2 and Proton Density) from the last
section that better represent tissues and it is
possible to say that those tissues with the
same color in each of these three images may be same tissues.
The combination of fuzzy systems and neural Networks make a powerful tool for pattern
recognition problems. So a fuzzified neural network with outputs to a back-propagation
network for tissues recognition must be used. Therefore a fuzzy neuron and a fuzzified
network are introduced. The output of the back-propagation network is the type of tissue
The results of the last section are fed into another network that uses a knowledge-base to
make a suggestion for treatment (This level is also optional).
Because of the time limitation MATLAB 4.0 for Windows is used for expert system
simulation. It has several abilities for matrix calculations and graphing that make the work
easier (some of the base modules are written in C++).
Fuzzy neural networks