ORIGINAL RESEARCH article
Front. Oncol.
Sec. Cancer Imaging and Image-directed Interventions
This article is part of the Research TopicInnovations in Cancer Imaging and Radiomics through Explainable Artificial IntelligenceView all 6 articles
EnsembleSkinNet: A Transfer Learning-Based Framework for Efficient Skin Cancer Detection with Explainable AI Integration
Provisionally accepted
- GITAM Deemed to be University, Hyderabad, India
Select one of your emails
You have multiple emails registered with Frontiers:
Notify me on publication
Please enter your email address:
If you already have an account, please login
You don't have a Frontiers account ? You can register here
Skin cancer remains to be one of the commonest and lifethreatening cancers in the world, early diagnosis is vital for an effective treatment. Recent advances in deep learning, specifically Convolutional Neural Networks (CNNs), have led to remarkable progress in skin lesion classification by artificial intelligence. Although single CNN-based methods have been shown to provide high accuracies on their specific datasets and patient conditions, variations in lesion morphology, image quality, and acquisition settings can limit the generalization of these methods on a new unseen dataset. To overcome these difficulties, we present EnsembleSkinNet which is more explainable ensemble deep learning framework for skin image classification based on a softmax-weighted spectrum fusion of different pre-trained CNN architectures including Modified VGG16 (M-VGG16), ResNet50, Inception V3 and DenseNet201. Our framework improves the robustness and reliability by using transfer learning, fine-tuning, and Bayesian hyperparameter optimization for classification. For experimental evaluation using five-fold cross-validation in HAM10000 dataset, it achieves an accuracy of 98.32 ± 0.41 %, precision = 98.20 ± 0.35 %, recall = 98.10 ± 0.38 %, and F1-score = 98.15 ± 0.37 %. Cross-domain generalization was also demonstrated by obtaining an excellent external validation accuracy of 96.84 ± 0.42 % and AUC = 0.983 on the ISIC 2020 dataset. Additionally, the Grad-CAM–based explainability analysis reached a mean Explainability Accuracy of 93.6 % (k = 0.87), indicating agreement with dermatologist annotations. Clinically, this means that false negatives for melanoma and basal cell carcinoma were measurable less, so more cases are caught early and diagnostically driven confidence improved. Hence, EnsembleSkinNet presents a simple, reproducible, interpretable, and clinically adopted framework for OA compliant robust AI-based skin cancer diagnosis.
Keywords: ensemble learning, skin cancer detection, deep learning, Convolutional neural networks (CNNs), Explainable AI, Transferlearning
Received: 11 Sep 2025; Accepted: 17 Nov 2025.
Copyright: © 2025 Cherukuri and Srisailapu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
* Correspondence: Srilakshmi Cherukuri, cherukurisreeit@gmail.com
Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.