A Deep Learning Approach for Monkeypox and Other Skin Lesion Classification

Monkeypox is a viral disease, originates from ani- mals and can spread to humans. It causes skin lesions that

often resemble those seen in conditions like chickenpox and cowpox, making accurate and timely diagnosis chal-

lenging. Accurately identifying monkeypox is essential because its distinctive skin lesions often resemble those

of other diseases, which can make diagnosis difficult and delay treatment. Often lack the precision needed to

tackle the specific challenges of diagnosing monkeypox and similar skin diseases, especially in resource-limited

settings. This study introduces to develop and validate an optimized AI driven framework for accurately classi-

fying monkeypox and related skin lesions, addressing the limitations of existing diagnostic systems by delivering

high accuracy, scalability, and applicability in real-world healthcare settings, especially in resource constrained

environments. Using Mpox Skin Lesion Dataset Version 2.0 enriched with extensive augmentation, and eight ad-

vanced deep learning models such as EfficientNetB5, ResNet50, ResNet101, MobileNet, Xception, DenseNet121,

NasNetMobile, and InceptionV3 were systematically evaluated through a rigorous 5-fold cross-validation, en-

suring robust generalization. The novelty lies in integrating adaptive augmentation, optimized preprocessing,

and lightweight architectures like EfficientNetB5 and MobileNet, designed for efficiency without compromising

accuracy. Both models outperformed others, achieving over (±90%) accuracy and F1 scores, demonstrating

exceptional efficiency and robustness for practical use. This framework addresses key limitations in automated

dermatological diagnostics by providing a scalable, precise, and resource-efficient solution, enabling healthcare

professionals to improve workflows, make informed decisions, and enhance patient outcomes, especially in re-

source limited and high-demand environments.