Analysis of Segmentation Method for Skin's Lesions Identification

Artificial Intelligence support for medical purposes has been raised in various aspects. The post pandemic issue stimulates the hype of tele-medicine technology support for rural areas. In the context of dermatology healthcare, the imbalance distribution of dermatologist makes general practitioners the main medical expertise who deliver dermatology healthcare in rural areas. Thus, tele-medicine occupied with AI-based diagnostic system will be very helpful to not only connecting the general practitioner to the dermatologist but also supporting them with AI-based skin diseases diagnostic system. Constructing AI model for skin diseases spot diagnostic system is not easy. Proving high performance accuracy should be delivered basedon given skin's diseases image. one of the ways to increase the performance is by localizing the lesion hence easy to capture the feature behind. Hence, image segmentation is urgent for lesion localization to improve the AI-diagnosis accuracy. This paper evaluates various approaches of segmentation method to localize skin's lesion on given image. The analysis is delivered subject to the skin pigmentation level, pigmentation contrast between the healthy skin and the lesion and the lesion distribution. As overall, Grab-cut show the most robust for any kind of given conditions. Even though the performance does not meet the requirements yet, the success rate is quite stable good on every condition with given constraints. Observation on semantic based segmentation method and instance-based segmentation is also discussed to address the localization objectives utilization.