Editorial
DermFusion-CBAMNet: A Dual-Backbone Attention-Guided Framework for Robust Multiclass Skin Disease Classification
@ARTICLE{10.4108/eetiot.13764, author={Padma Jyothi Uppalapati and Sridevi Bonthu and Leelavathi Arepalli}, title={DermFusion-CBAMNet: A Dual-Backbone Attention-Guided Framework for Robust Multiclass Skin Disease Classification}, journal={EAI Endorsed Transactions on Internet of Things}, volume={11}, number={1}, publisher={EAI}, journal_a={IOT}, year={2026}, month={10}, keywords={Skin disease, Transfer Learning, Fusion model, EfficientNet B2, DenseNet121, VGG19}, doi={10.4108/eetiot.13764} }- Padma Jyothi Uppalapati
Sridevi Bonthu
Leelavathi Arepalli
Year: 2026
DermFusion-CBAMNet: A Dual-Backbone Attention-Guided Framework for Robust Multiclass Skin Disease Classification
IOT
EAI
DOI: 10.4108/eetiot.13764
Abstract
Skin diseases are a major health care problem all over the world. Accurate and prompt diagnosis of skin disorders can result in efficient treatment and better patient outcomes. However, various skin disorders still remain challenging to be automatically classified due to the significant inter-class similarities, intra-class variances, diversified lesion appearances and complicated image backgrounds. We propose DermFusion-CBAMNet, a new dual-backbone attention-based deep learning architecture for multiclass skin disease categorization. Instead of the typical transfer learning techniques that use only one pre-trained model or just replace the classification layers, the framework proposed here maintains the structural integrity of two complementary pre-trained architectures, DenseNet121 and EfficientNet-B2, and merges their learned features through an efficient feature fusion strategy. In addition, both feature extraction branches are integrated with Convolutional Block Attention Module (CBAM) to refine the channel-wise and spatial feature representations by focusing on diagnostically relevant lesion characteristics selectively and suppressing redundant information. Such attention-guided fusion helps the model to acquire more discriminative and robust representations to correctly classify several kinds of skin diseases. The proposed architecture is assessed using a benchmark dataset that contains 10 types of skin illness, with the help of comprehensive performance metrics like accuracy, precision, recall, F1-score, Receiver Operating Characteristic (ROC) and Area Under the Curve (AUC). The experimental findings show that DermFusion-CBAMNet beats the baseline models. These results prove that DermFusion-CBAMNet is a robust, interpretable and reliable framework for automated multiclass skin disease classification and has great potential to be deployed in computer-aided dermatological diagnosis and clinical decision support systems.
Copyright © 2026 P. Uppalapati et al., licensed to EAI. This is an open access article distributed under the terms of the CC BY-NCSA 4.0, which permits copying, redistributing, remixing, transformation, and building upon the material in any medium so long as the original work is properly cited.