Virtual Booth

ABSTRACT

Introduction: Skin cancer arises from impaired skin regeneration. Melanoma is one of the most dangerous skin malignancies and causes many death. Without early screening and diagnosis, preventive measures will be very difficult and may increase morbidity and mortality. Populations in tropical, coastal areas like Semarang City face high risks due to prolonged UV exposure. Medical data (August-December 2024) from a Semarang hospital shows patients are predominantly females aged ≥40, with non-melanoma skin cancers frequently appearing on the head and neck. However, public health threats also include 30 contagious skin diseases. Because manual diagnosis is time-consuming, automated solutions are essential.

Objectives: To develop an EfficientNet AI model using PyTorch, integrated into the MelanoCare mobile application for early, comprehensive skin disease detection.

Methods: Skin images undergo preprocessing with black-hat filtering and digital inpainting to remove hair artifacts. The EfficientNet model is then trained to classify melanoma, non-melanoma cancers, and 30 contagious skin conditions.

Results: It is expected that the results of our study will facilitate accurate independent screening. The EfficientNet model is anticipated to handle class imbalance effectively, achieving high F1-scores. Its lightweight nature ensures seamless real-time performance within the MelanoCare app.

Conclusion: Combining PyTorch and EfficientNet delivers a smart, fast, and practical early skin disease screening tool for daily public use.

Keywords: Skin Cancer, Contagious Diseases, PyTorch, EfficientNet, MelanoCare