Abstract:
Skin lesions represent a widespread and serious global health concern. Timely detection of
melanoma in images from dermoscopy significantly increases the chance of survival.
Accurately diagnosing melanoma presents several challenges, such as low contrast between
lesions and healthy skin, and visual similarities between melanoma and non-melanoma lesions.
This challenge highlights the necessity for trustworthy automatic detection techniques to
improve dermatologists’ precision and productivity. In recent years, there has been a notable
surge in attention toward deep learning techniques for image analysis. These techniques,
leveraging machine learning, transform input data into high-level representations. Deep
learning networks, with their ability to perform segmentation, utilize conventional filters, and
incorporate pooling layers, have gained particular interest in the medical field for accurate
diagnoses, especially in melanoma detection. In response to these challenges, this project
proposes deep learning-based solutions for skin lesion analysis, specifically focusing on
dermoscopic images containing skin cancer. The models are trained and evaluated on a standard
benchmark dataset, demonstrating promising accuracy. The solution is to build a mobile
application for skin cancer detection that utilizes artificial intelligence and image processing
technologies. One of part of the presented work also includes the creation of the prototype of
a melanoma skin cancer detection mobile application aimed at the identification of skin
lesions which will automatically provide feedback, and outcomes as well as maintain a
history of skin images, including those captured by dermatologists. It allows the user to do
self-checks and get an alert on possible skin lesions or cancer. It focuses on the effective
diagnosis of skin lesions by using models that help increase the degree of reliability thereby
encouraging early diagnosis of skin health and probably preventing loss of lives.
