Identifying melanoma characteristics using directional imaging algorithm and convolutional neural network on dermoscopic images / Mohammad Asaduzzaman Rasel

Mohammad Asaduzzaman , Rasel (2024) Identifying melanoma characteristics using directional imaging algorithm and convolutional neural network on dermoscopic images / Mohammad Asaduzzaman Rasel. PhD thesis, Universiti Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (220Kb)
	PDF (Thesis PhD) Download (4Mb)

Abstract

Melanoma is the deadliest skin cancer worldwide. Advancements in digital dermoscopic image analysis have greatly improved computer-aided Melanoma diagnosis systems. The use of dermoscopic images early detection of Melanoma has gained popularity among researchers due to its non-invasive nature. This thesis aims to enhance the analysis of dermoscopic images for identifying Melanoma. A critical first step for this is to distinguish between healthy and unhealthy skin areas by improving the segmentation process. This is followed by lesion features extraction and analysis (including classification) based on clinically diagnosis criteria including ABCDE rules, 3-point checklist, 7-point checklist, and CASH, to automate the manual process. This research is divided into two phases – 1) Feature Engineering phase explains skin conditions based on lesion segmentation and different dermoscopic feature extraction, while 2) Classification phase detects Melanoma. Multiple deep-learning models are proposed for segmentation. Several imaging, computer vision, and pattern recognition algorithms are employed to describe five dermoscopic features. Subsequently, these features are classified using different proposed deep learning models on various publicly available datasets. To overcome the issues with non-annotated dataset, several techniques are proposed. Both phases of the research outputs are evaluated and compared with the state-of-the-art methods. The proposed algorithms that outperformed the state-of-the-art algorithms contributes to diagnosing early-stage Melanoma. Findings from this study would help dermatologists and patients reduce the time and cost of Melanoma diagnosis, while receiving explanation for such automated diagnosis. Only five most common and important features of many Melanoma-features are analyzed. As part of future work, incorporating more dermoscopic features such as irregular blotches and regression structures in the analytical section would be promising.

Actions (For repository staff only : Login required)