1*School of Computer Engineering, Kalinga Institute of Industrial Technology, Bhubaneswar, Odisha, India. Email: abhipsaapattanaik@gmail.com. ORCID ID: 0009-0009-0193-6752
2School of Computer Engineering, Kalinga Institute of Industrial Technology, Bhubaneswar, Odisha, India. Email: ldasfcs@kiit.ac.in. ORCID ID: 0000-0002-8719-1991
Context: Skin cancer is one of the most dangerous diseases in the world. Diagnosing it with a traditional model struggles to give accurate results and suffers from computational efficiency and clinical interpretability. But handling such diseases through deep learning gives promising results nowadays.
Objective: In this paper, we aim to classify skin disease using deep learning models. We propose a deep learning model called FusionSegNet that combines the benefits of Transformers and CNNs to provide clinically reliable and highly accurate skin disease classification.
Material/Methods: We used several deep learning models to integrate to detect the skin disease. Our method combines multi-scale feature extraction from the ResNet50 and Vision Transformer (ViT) branches with attention-guided fusion to identify diagnostically significant patterns. On the ISIC 2018 dataset, FusionSegNet outperforms cutting-edge techniques like ConvNeXt (92.3%) and DeepSkinNet (92.1%) with an accuracy of 96.3%.
Result: We used nine deep learning models, such as MobileNetV3, EfficientNet-B4, ResNet50, DenseNet121, ViT-B/16, DeepLabV3+, Swin-Tiny, DeepSkinNet (SOTA), ConvNeXt-Small, and our proposed model, FusionSegNet. Our proposed hybrid deep learning model obtained the highest accuracy of 96.3%, precision of 95%, recall of 96%, F1-score of 95% AUC of 98% as comparisons to the other models. We also used the ablation and statistical significance to confirm that each architectural component shows that the model's performance is increasing. FusionSegNet is a suitable solution for reliable, efficient, and generalizable skin disease diagnosis, with a high potential for integration into clinical decision-support systems.
Keywords: Skin cancer classification; deep learning, multi-model fusion; explainable AI; medical image analysis.
How to cite this article: Pattanaik A, Das L. FUSIONSEGNET: A Deep Learning Framework For Accurate And Explainable Skin Disease Classification Using Multi-Model Integration. Int J Drug Deliv Technol. 2026;16(16s): 333-354. DOI: 10.25258/ijddt.16.16s.36
Source of support: Nil.
Conflict of interest: None