Advanced biomedical imaging for plaque lesion diagnosis focusing on Malaysian subjects / Hadzli Hashim

This thesis describes the development of a novel non-invasive color based intelligent diagnosis model for plaque psoriasis lesion focusing on Malaysian subjects. The system which is based on primary color components from digital images employed new algorithms of data acquisition, data processing, data extraction and application of artificial neural network (ANN) as the decision model to discriminate plaque from other major psoriasis. Two major works were carried out; one was the extraction process of a single color component for plaque discrimination through application of known statistical tools on the clustering pixel gradation indices of each color component in terms of location and shape. Second part of the work was to design the ANN diagnosis models by utilizing the extracted single color spectrum with various combinations of its gradation indices. A multi color spectrum ANN model, where it used all the three primary components was designed and treated as a controlled system. These models were evaluated and validated through analysis of the performance indicators applied in medical research; sensitivity, specificity, clustering properties and discriminative power of the models by plotting the effects of threshold adjustment on their diagnostic accuracy, error and uncertainty (DA, DE and DU), and the optimum Euclidean Distance (ED) from the ideal point (1,0) in the receiver operating characteristics (ROC) plot. Other than that, the respective model’s network structure was also considered.