Amr Sharawy

Automatic Detection of Melanoma Skin Cancer using Texture Analysis

Melanoma is considered the most dangerous type of skin cancer. Early and accurate diagnosis depends mainly on important issues, accuracy of feature extracted and efficiency of classifier method. This paper presents an automated method for melanoma diagnosis applied on a set of dermoscopy images. Features extracted are based on gray level Co-occurrence matrix (GLCM) and Using Multilayer perceptron classifier (MLP) to classify between Melanocytic Nevi and Malignant melanoma. MLP classifier was proposed with two different techniques in training and testing process: Automatic MLP and Traditional MLP. Results indicated that texture analysis is a useful method for discrimination of melanocytic skin tumors with high accuracy. The first technique, Automatic iteration counter is faster but the second one, Default iteration counter gives a better accuracy, which is 100 % for the training set and 92 % for the test set.

K6. Circular binary segmentation modeling of array CGH data on hepatocellular carcinoma

Hepatocellular carcinoma (HCC) is a malignant tumor derived from hepatocytes that belong to primary malignant epithelial tumors of the liver. The outcome of HCC patients still remains dismal due to the difficulty in detecting the disease at its early stage. We propose a new approach aiming to identify new biomarkers for early diagnosis of HCC. Genomic DNA copy number alterations (CNAs) are associated with complex diseases like HCC. Array-based Comparative Genomic Hybridization (a-CGH) is a technique used to identify copy number changes in genomic DNA. We use a statistical model based on a circular binary segmentation (CBS) algorithm. Our approach makes use of a median absolute deviation model to separate outliers from their surrounding segments. We tested 35 samples of HCC patients on specific chromosome regions, then applied CBS algorithm to detect genomic DNA alternations in copy number. Our results show that a gain of 1q was detected in 63% and a gain of 20q was detected in 26% of HCC cases. Also, a loss of 4q was detected in 3%, a loss of 13q was detected in 29%, loss in 16q was detected in 9%, and loss of 17q was detected in 3% of HCC cases.

A New Method to Grid Noisy cDNA Microarray Images Utilizing Denoising Techniques

DNA Microarray is an innovative tool for gene studies in biomedical research, and its applications can vary from cancer diagnosis to human identification. It is capable of testing and extracting the expression of large number of genes in parallel. The gene expression process is divided into three basic steps: gridding, segmentation, and quantification. Automatic gridding; which is to assign coordinates to every element of the spot array, is considered the most challenging phase of microarrays image processing. For processing of microarray images, a new, automatic, fast and accurate approach is proposed for gridding noisy cDNA microarray images. In the real world, microarray image doesn’t reflect measures of the fluorescence intensities for the dye of interest only, as different kinds of noise and artifacts can be observed. In this paper, a novel gridding method based on projection is developed accompanied by a pre-processing, post-processing, and refinement steps for noisy microarray images. Results revealed that the proposed method is used with high accuracy and minimal processing time and can be applied to various types of noisy microarray images.

Pigmented Skin Lesion Diagnosis by Automated Imaging System

Pigmented skin lesions are the normal part of the skin, however its anomalous appearance is an annoying sign due to the presence of melanomas one of its malignant forms. Although melanoma is a deadly considerable disease, its early detection is a serious step toward mortality reduction. The proposed research discusses the outcome of introducing312 different features in a non-invasive diagnosis method for pigmented skin lesion diagnosis. To obviate the problem of qualitative interpretation, two different image sets are utilized to examine the proposed system, a set of images acquired by standard camera (clinical images) and another set of dermoscopic images captured from the magnified dermoscope. System contribution appears in using large conclusive set of features fed to different classifiers composing totally complete, new and different approaches for the purpose of disease diagnosis. Miscellaneous types of features used such as geometric, chromatic, and texture features extracted from the region of interest resulted from segmentation process. Then, the most prominent features that can cause an effect are selected by three different methods; Fisher score method, t-test, and F-test. The high-ranking features are used for the diagnosis of the two lesion groups using Artificial Neural Network (ANN), K-Nearest Neighbor (KNN) and Support Vector Machine (SVM) as three different classifiers proposed. System performance was measured in regards Specificity, Sensitivity and Accuracy. The ANN designed with the feature selected according to fisher score method enables a diagnostic accuracy of 96. 25% and 97% for dermoscopic and clinical images respectively.

Ultrasound Transducer Quality Control and Performance Evaluation Using Image Metrics

This paper aims to two main goals, first goal is to achieve the characterization of quality control of ultrasound scanners based on the potential image metrics. On the other hand, the most effective goal is how to classify ultrasound scanners based on image metrics to evaluate performance of ultrasound transducer. The authors utilize the metrics to give information about the spatial arrangement of the gray levels in the specific interest region. The execution of ultrasound images metric based on a set of 19 metrics (i.e. contrast, gradient and Laplacian). This set reflects quality control of ultrasound scanners. The wok of this paper based on the best 6 metrics from 19 metrics which extracted from linear discriminative analysis (LDA). The classification methods used for minimum numbers of metrics are fused using support vector machine (SVM) and the highest classification method is back propagation neural network (BPNN) classifiers to get the main target of paper. Finally, the results show that objective performance evaluation of ultrasound transducer accuracy was 100% by using back propagation neural network classifier.

Sensitivity of the Risk Factors for the Progression of Ocular Hypertension to Primary Open Angle Glaucoma

In this article we evaluate the sensitivity of the risk factors of ocular hypertension progression in primary open angle glaucoma in order to distinguish between the three risk levels based on prediction classification models. The prediction classification models were trained and testing by using the most common risk factors from examination of 398 Egyptian patients. Standard classification trees as well as bagged classification were used. We classify the risk level into three risk levels which are high, middle and low based on the combination of the structural and functional risk factors. The classification outcomes of the trees were compared and we measured the sensitivity of each risk factor. The bagged classification has the best accuracy which is 87.7% for training datasets and 72.2% for testing datasets with area under the receiver operating characteristics curve (AUROC) 0.925 while decision tree gave 80% for training datasets and 68.7% for testing datasets with AUROC 0.868. The central cornea thickness (CCT) gave the best with average AUROC 0.946. Bagged classification tree promises to be a new and efficient approach for glaucoma classification. The CCT is very important risk factor due to its classification sensitivity.