Early Diagnosis and Predictive Monitoring of Skin Diseases

Abstract

Skin diseases like melanoma are traditionally screened by a visual analysis of key features, such as the pigmentation and vascularity of the region of interest. Monitoring the changes of these features during follow-up imaging sessions is critical towards a correct medical diagnosis. This paper proposes a framework to monitor these changes on the skin over time. The proposed framework utilizes the Lucas-Kanade displacement flow implementation to detect the severity of spatial changes in the skin. These spatial changes are captured via the magnitude and direction of the vectors in the resultant displacement field. This change monitoring is tested for surface and sub-surface skin image data. The proposed framework is developed and validated with skin samples of cutaneous melanoma, setting the stage for future extension to images from other skin diseases. The skin sample images would be classified as high and low risk based on the severity of change. Further, a predictive algorithm is devised to estimate the change in the high risk skin lesions, providing significant information to determine the course of medical action.