| Περίληψη: | Melanoma is one of the deadliest forms of skin cancer. Cases of melanoma have increased significantly over the last decades. For young adults, it is estimated that is one of the most often diagnosed cancers. Nevertheless, there are many treatment options if detected early. Unfortunately, current diagnostic techniques for screening patients for melanoma are too expensive. In this context, an automated approach that uses photos of a patient's skin lesions to assist the diagnosis of melanoma is needed. Dermatologists may utilize this technology to diagnose patients without having to invest in pricey or specialized equipment.
This thesis aim is to develop a framework to segment skin lesion images from photographs that have been taken by a common digital camera. Three segmentation techniques are employed in this project: two semi-automatic methods and one automatic. The semi-automatic methods are provided by image processing software, and the proposed automatic segmentation method is developed for this project. The automatic method contains a simple segmentation program and extension algorithms for application in segmented images with low accuracy. Each method is performed in a dataset without previous segmentation and is tested by comparing skin lesions from segmented images against ground truth segmentation, which is provided by a specialist medical physicist with the assent of an expert dermatologist. The performance of each method is evaluated with metrics, such as specificity.
|
|---|
