الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 168 |  |  |
| | | from | 168 | | |
Abstract
Medical diagnostics is a tool that assists physicians in understanding the behavior of the human body’s internal organs. Medical datasets are gathered from various resources with different characteristics, spatial domain, time-frequency domain, and medical images. The spatial domain datasets have case sensitive data. Moreover, they suffer from feature correlation, data dependency, and feature redundancy problems. Although several methods have solved these issues, the time complexity of these methods is not satisfactory. In addition, the problem of correlated features preventing some significant features from participating in knowledge reduction is still under consideration. Moreover, feature extraction and data decomposition are other obstacles in handling the time-frequency domain dataset. The time-frequency domain requires much processing power because of the huge amounts of data gathered from various levels of decomposition. Finally, medical images are captured in a non- uniform, low-luminance, and low-contrast way because of the real-world restrictions of the capturing process. In this dissertation, a novel approach is proposed based on mutual information and a modified firefly optimization algorithm to train the cellular model and find the optimal values of its templates. For spatial domain datasets, a new uncertainty measure is defined based on mutual information with a non-uniformity estimator that determines the significance of each attribute. The monotonicity of the measure is mathematically proved. For the time frequency domain, a modified firefly algorithm was designed with a data fusion technique to navigate the search space. The modified firefly was combined with a tunable wavelet to classify medical data for epilepsy. For medical images, a cellular neural network is trained using the uncertainty measure and the firefly to remove the noise. The uncertainty measure was utilized to detect the similarities among the cells and remove the superfluous and redundant cells from the template. The obtained templates are examined on the X-ray image to report the significance of the templates. As a conclusion, an uncertainty measure was proposed with minimized time complexity to determine the significance of a specific feature. Also, a modified firefly optimization algorithm was presented with a minimized attraction model. An innovative learning method based on the proposed measure and the modified firefly was proposed to train the cellular neural network. The experimental results showed the significance of the proposed approach when compared to different solutions.