Please use this identifier to cite or link to this item:
|Title||Melanoma Detection Using Mobile Technology and Feature-Based Classification Techniques|
Melanoma is one of the most dangerous types of skin cancer in terms of the ratio of death cases. Probability of death increases when it is diagnosed late. However, it is possible to treat melanoma successfully when diagnosed in its early stages. One of the most common medical methods for diagnosing melanoma is the ABCD (Asymmetry, Border irregularity, Color, and Diameter) method that involves the measurement of four features of skin lesions. The main disadvantage of this method is that estimation error and subjectivity affects the accuracy of diagnosis, especially when performed by non-specialists. Scarcity of specialists makes the problem worse. This has led to the development of computer systems to help in melanoma diagnosis. However, while most computer systems can achieve high accuracy with adequate speed, they have problems in the usability and flexibility. The emergence of smart phones with increasing image capture and processing capabilities has made it more possible to use such devices to perform medical image analysis such as the diagnosis of melanoma. Our research work combines existing melanoma diagnosis method and the image capture and processing capabilities of smart phones to achieve fast, affordable, easily available and highly accurate melanoma diagnosis. In this work, we propose a complete smart phone application to capture, and process an image of the suspicious region of the skin in order to estimate its probability of being melanoma. The system can use historical cases to improve its diagnosis accuracy. The system was tested on 164 sample images. 14 images were not well-captured and could not be diagnosed, while the remaining 150 cases were successfully processed. In each of these 150 images, the lesion was correctly segmented and their ABCD feature set extracted. Diagnosis accuracy of the analyzed images ranged between 88%-94 with best results using SVM classifier, and worst is the KNN classifier.
|Publisher||الجامعة الإسلامية - غزة|
|Files in this item|