Why not?! But you need very clear ideas, well-defined concept and tons of literature. And don't forget that the appearance of melanoma (in lesser extent of other skin tumors) is quite heterogeneous. All types of algorithms rely on what is typical for melanoma, and makes the difference between benign and malignant lesions. In real-life the main problem of early melanoma detection are those melanomas appearing without typical features. It is extremely difficult, if not impossible, to implement such cases in an algorithm.