SkinFusionNet: MobileNetV2 和 Vision Transformer 结合的深度学习模型用于皮肤癌分类
SkinFusionNet: MobileNetV2 and vision transformer hybrid deep learning model with automated skin cancer classification
该研究提出了一种名为 SkinFusionNet (SFN) 的新型混合深度学习模型,结合了 MobileNetV2 和 Vision Transformer 的优势,用于基于皮肤镜图像的皮肤癌自动分类。
- Article
- Open access
- Published: 09 November 2026
- Hafiz Muhammad Zeeshan1,
- Muhammad Aksam Iftikhar2,
- Attiq Ur Rehman3,
- Sohail Masood Bhatti1,4,
- Arfan Jaffar1,4,
- Abdullah Y. Muaad5 &
- …
- Md Belal Bin Heyat6
Scientific Reports (2026) Cite this article
We’re sharing this article early to provide faster access to peer-reviewed, accepted research. It is citable and carries a permanent DOI. This version is subject to further edits and will be replaced automatically by the final Version of Record. All legal disclaimers apply.
Abstract
Skin cancer is one of the most widespread and potentially fatal diseases worldwide, with millions of new cases diagnosed every year. Detecting malignant skin lesions early and accurately is crucial for improving patient survival and can also help reduce the overall cost of treatment. Nevertheless, the conventional methods of diagnosing are usually time-consuming, subjective, and dependent on the skills of a clinician, thus not providing homogeneous results. In order to solve such shortcomings, this paper presents an original hybrid deep learning model, SkinFusionNet (SFN), as an automated skin lesion classifier based on dermoscopic images. The suggested SFN model is a synergetic integration of the advantages of MobileNetV2 and Vision Transformer (ViT) architectures the ability to extract local features efficiently by using MobileNetV2 and the ability to learn long-range contextual dependencies by ViT. The manipulation of both fine-grained lesion textures and larger image structures by such integration results in the high diagnostic ability of SFN at a wide range of imaging settings. The model was trained and tested on the 2018 ISIC dataset which had been augmented and balanced to equalize malignant and benign samples at 3,000 and 3,000, respectively. With SFN, outstanding performance was observed with 95.44 per cent accuracy, 95.48 per cent precision, 95.44 per cent recall and a Dice score of 0.9544, which is high compared to current CNN based and hybrid models like ResNet, DenseNet and MobileNet variants. Grad-CAM visualizations were used to mark important areas of the images that had an impact on model predictions to increase clinical trust and interpretability to offer important information about the decision-making process. The findings have shown that SFN can be an effective, interpretable and high-performance framework used in the initial detection of skin cancer. It can be easily deployed in the real world due to its lightweight nature and good ability to generalize, especially in healthcare settings that have limited resources. The given strategy is an important move on the way to making explainable AI a part of dermatological diagnostics, which would make the process of skin cancer screening faster, more credible, and more accessible.
Abbreviations
- AI:
-
Artificial intelligence
- AKIEC:
-
Actinic keratosis / intraepithelial carcinoma
- BCC:
-
Basal cell carcinoma
- BKL:
-
Benign keratosis-like lesions
- CNN:
-
Convolutional neural network
- DF:
-
Dermatofibroma
- DL:
-
Deep learning
- DT:
-
Decision tree
- FN:
-
False negative
- FP:
-
False positive
- Grad-CAM:
-
Gradient-weighted class activation mapping
- ISIC:
-
International skin imaging collaboration
- KNN:
-
K-nearest neighbor
- MEL:
-
Melanoma
- ML:
-
Machine learning
- NB:
-
Naive bayes
- NV:
-
Nevus
- SFN:
-
SkinFusionNet
- SVM:
-
Support vector machine
- SCC:
-
Squamous cell carcinoma
- TN:
-
True negative
- TP:
-
True positive
- TL:
-
Transfer learning
- ViT:
-
Vision transformer
- VASC:
-
Vascular lesion
- XAI:
-
Explainable artificial intelligence
Acknowledgements
The authors would like to acknowledge the International Skin Imaging Collaboration (ISIC) for providing the publicly available ISIC 2018 dermoscopic image dataset, which served as the primary data source for this research. The availability of this dataset has significantly contributed to the development and evaluation of automated skin lesion classification systems. The authors also appreciate the support of their respective institutions for providing the academic and computational resources necessary to conduct this research. In addition, the authors are grateful to colleagues and researchers who provided valuable discussions and feedback during the preparation of this study.
Funding
None.
Author information
Authors and Affiliations
Department of Computer Science, The Superior University, Lahore, Pakistan
Hafiz Muhammad Zeeshan, Sohail Masood Bhatti & Arfan Jaffar
Department of Computer Science, COMSATS University Islamabad, Lahore Campus, Lahore, Pakistan
Muhammad Aksam Iftikhar
Department of Cyber Security, School of Cyber Science and Engineering, Huazhong University of Science and Technology, Wuhan, China
Attiq Ur Rehman
Intelligent Data Visual Computing Research, Lahore, Pakistan
Sohail Masood Bhatti & Arfan Jaffar
IT Department, Sana’a Community College, Sana’a, Yemen
Abdullah Y. Muaad
School of Engineering, Westlake University, Hangzhou, China
Md Belal Bin Heyat
Authors
- Hafiz Muhammad Zeeshan
- Muhammad Aksam Iftikhar
- Attiq Ur Rehman
- Sohail Masood Bhatti
- Arfan Jaffar
- Abdullah Y. Muaad
- Md Belal Bin Heyat
Corresponding authors
Correspondence to Abdullah Y. Muaad or Md Belal Bin Heyat.
Ethics declarations
Competing interests
The authors have no competing interests.
Additional information
Publisher's note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
About this article
Cite this article
Zeeshan, H.M., Iftikhar, M.A., Rehman, A.U. et al. SkinFusionNet: MobileNetV2 and vision transformer hybrid deep learning model with automated skin cancer classification. Sci Rep (2026). https://doi.org/10.1038/s41598-026-74475-7
Received: 28 March 2026
Accepted: 29 September 2026
Published: 09 November 2026
DOI: https://doi.org/10.1038/s41598-026-74475-7
Keywords
来源:Nature 健康护理主题 · nature.com