CNN Architectures Exploration and Analysis for Multiclass Skin Diseases Identification

Skin diseases have become one of the most common ailments, particularly in rural areas. Misdiagnosis, however, can lead to an even worse quality of life and inefficient treatments. Furthermore, countries with archipelagos, like Indonesia, are particularly impacted by an imbalance in the distribution of dermatological care. Here, general practitioners with less experience in the field typically provide most dermatological care in this area. Thus, misdiagnosis is virtually inevitable and leads to problems for the patient, even impacting the doctor's and dermatology organization's reputation. To solve this problem, it has recently been shown that general practitioners can identify skin disorders using artificial intelligence (AI). Unfortunately, not all AI implemented has considered dermatologists' knowledge in their systems, and specifically, most Indonesian dermatologists are not familiar with this AI thing. Thus, this work was purposed to help them predict skin disease classification through CNN-based Deep Learning Architecture methodologies for autonomous skin disease identification using mobile photos and dermoscopy. Here, the Fitzpatrick 17k and ISIC 2018 datasets with several deep learning methods, including ensemble techniques were used. The results showed that for mobile-based skin disease photos, the deep learning architecture offers a notable improvement over the handcrafted-based approach. The accuracy using an ensemble for the Fitzpatrick dataset was 55% and ISIC was higher, 80%. Although not all methods are improving slowly, further employment and refinement of ensemble techniques promise to automate further medical diagnostics which can enhance patient outcomes in dermatology. Finally, additional images will increase accuracy; specifically with the more intricate the photographs and sizes. Thus, this preliminary research showed the importance of the deep learning method in identifying skin disease classification.