Basic Informations

C.V

Name: hagar moahmed reda ali
Date of Birth and Place: 1/1/1988- Beni Seuif
Nationaly: Egyptian
Marital status: Single
Address: Faisal, Giza

Master Title

Data mining techniques for medical diagnosis

Master Abstract

ABSTRACT The successful application of data mining in highly visible fields like e-business and marketing have led to the popularity of its use in knowledge discovery in databases (KDD) in other industries and sectors. Among these sectors that are just discovering data mining are the fields of medicine and public health. The medical industries collect huge amounts of healthcare data which, unfortunately, are not “mined” to discover hidden information. We can describe this data as being ‘information rich’ yet ‘knowledge poor’. In this work, we briefly examine the use of the most important data mining techniques such as Artificial Neural Network to massive volume of data in medical field which is pediatric respiratory disease. After we had trained the network with 699 cases contains input vector and target vector, we tested it with 20 cases without the target vector. The output of the tested cases is one of the following eight examinations (bronchiolitis, pneumonia, acute epiglottitis, pleurisy, emphysema, acute laryngotracheobronchitis, bronchial asthma and bronchiectasis). The present data explained that 90% of all test cases represent the correct examination. This means that the experimental results on this medical data illustrate that neural networks are important in Diagnosis of medical data, especially for a large amount of data in a high-dimensional space. Also we design an intelligent algorithm for automatic recognition of pediatric respiratory diseases. Data mining Cluster analysis has been widely used in several disciplines, such as statistics, software engineering, biology, psychology and other social sciences, in order to identify natural groups in large amounts of data. In addition we briefly examine the implementation of both Principle Component Technique and Self Organization map (SOM) clustering techniques to massive volume of medical data in one of medical field which is pediatric respiratory disease. Then we compare results of SOM cluster technique with the result of Principal Components Analysis cluster process (PCA). The simulation results show that SOM performs better than PCA recognizing.

PHD Title

Cattle Identification Using Bovines Muzzle Patterns

PHD Abstract

ABSTRACT Ministry of Agriculture (Livestock Sector) and veterinarians search for new artificial techniques to save bovine's livestock and products. Bovines muzzle images are considered a biometric identifier to guarantee and maintain the safety of bovine's livestock products. This thesis presents nine different experiments to identify bovines depending on their muzzle images. Each experiment has three different parts. The first experiment depends on two parts before the last part which is the identification part using Artificial Neural Network (ANN). The pre-processing part consists of two phases; histogram equalization and mathematical morphology while the feature extraction uses box-counting algorithm. The second experiment is based on using histogram equalization and mathematical morphology filter in the first part, the second part is based on Segmentation-based Fractal Texture Analysis (SFTA) instead of box-counting algorithm and the last part uses ANN. The third experiment is based on histogram equalization and mathematical morphology filter in the first part and texture feature extracting part is based on box-counting and SFTA. The fourth one is based on using histogram equalization and mathematical morphology filter in the first part, box-counting algorithm in the second part and decision tree in the last part. The fifth experiment replaces the second part in the fourth experiment with SFTA and all phases still as it. The sixth experiment used average filter and median filter in the first part in order to remove noise from muzzle image and save the original pixels of the muzzle image, the second part is based on using Gray level co-occurrence matrix (GLCM) in order to extract muzzle features for each bovine, the last part is based on using Naïve Bayes. The seventh experiment is as the previous experiment. The only difference is in the last phase which is identification part and uses decision tree. The eighth experiment likes the sixth one but the difference in the second part which uses Discrete Wavelet Transform (DWT). The ninth and last experiment is like the seventh one but the second part uses Discrete Wavelet Transform (DWT) in order to extract image texture feature vector. The first five proposed experiment compare between the accuracy rate that extracted from box-counting, Segmentation-based Fractal Texture Analysis (SFTA) and the combination between them for texture feature extraction. Box-counting algorithm gives a texture feature vector consists of eight features and SFTA gives eighteen features for each bovines muzzle images. The combination features are twenty-six features for each bovine that we use in the third experiment. The following four experiments compare between the accuracy rate extracted from Gray Level Co-occurrence Matrix (GLCM) and Discrete Wavelet Transform (DWT) used in the second part which is feature extracting part. GLCM feature vector consists of twenty-two different feature and DWT feature vector consists of sixteen different features in the feature vector. The experimental results show the outperformance of Decision Tree classifier on ANN, and Naive Bayes. It declares that the best results are achieved in the seventh experiment which depends on texture feature extraction using GLCM and Decision Tree classifier. ANN achieves high accuracy rate when number of different bovines groups are five and three and the feature extracted using SFTA algorithm. In the second experiment; more accurate results achieved when we use five and three different classes. The accuracy also depended on the number of features that extracted in feature extraction part.