Title : Challenges in the development of computer-Aided diagnosis methodologies for skin lesion classification
Abstract:
The early detection of skin diseases is one of the priority tasks today worldwide. In 2013, skin diseases were the fourth leading cause of non-fatal illnesses that caused economic loss due to disability. Among the main risk factors are ultraviolet radiation and tanning beds. The diagnosis of skin lesions depends significantly on the expert's clinical experience in the area. It has been measured that the range of precision in specialized dermatological centers is between 64% to 80%. Therefore, computer-aided diagnosis (CAD) is beneficial to dermatologists and their assistants to make diagnoses more objective. However, there are significant and varied challenges to developing such applications. Among them are that skin lesions have irregular edges, non-uniform illumination, low contrast between healthy skin and lesion, databases with a different number of images by class, and in some cases, lack of representativeness in the type of lesions, in addition to the fact that the images are of varying size. On the other hand, the computer-aided diagnosis should be reproducible and robust; then, computational experiments must be designed in such a way that there are no biases, so it is vital to report performance metrics for the training set as well as the test set. And, at least the CAD's performance metrics to present should be accuracy, precision, sensitivity, and specificity.
