ABSTRACT:
ackground: Melanoma accounts for approximately 1% of all skin cancers but causes over 80% of related deaths worldwide. Early and accurate detection significantly improves survival; however, limited access to dermatologists, especially in low- and middle-income countries, restricts timely diagnosis. Deep learning, a subset of artificial intelligence (AI), has shown potential in improving diagnostic accuracy using medical imaging data.

OBJECTIVE:
This study aims to develop and evaluate a deep learning model capable of detecting melanoma from dermoscopy images to support early diagnosis and bridge gaps in dermatologic care.

METHODS:
A novel deep learning model, MelanomaENV2, was created using the EfficientNet V2 architecture. A total of 919 histopathologically confirmed dermoscopy images (400 melanoma and 519 nonmelanoma) from the HAM10000 dataset were preprocessed and divided into training (80%) and testing (20%) subsets. Transfer learning was applied using TensorFlow with binary cross-entropy loss, sigmoid activation, and a 20% dropout rate. Model performance was assessed using precision, recall, F1 score, accuracy, and AUC-ROC metrics.

RESULTS:
Among eight EfficientNet V2 variants tested, efficientnetv2-l-21k-ft1k achieved the highest performance: precision 0.73, recall 0.69, F1 score 0.71, test accuracy 0.72, and AUC-ROC 0.72. These results demonstrate promising discriminative capacity for classifying melanoma versus nonmelanoma lesions.

CONCLUSIONS:
The MelanomaENV2 model demonstrates the feasibility of using AI-based deep learning for melanoma detection in dermoscopy images. With further refinement and validation, it may serve as a valuable triage tool in primary and community healthcare, enhancing early diagnosis where dermatologic expertise is limited.