ISSN 2079-3537      

 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                                                                                             





Scientific Visualization, 2026, volume 18, number 3, pages 123 - 134, DOI: 10.26583/sv.18.3.09

Robust Monkeypox Detection Using Hybrid Deep Learning and Feature Fusion Techniques in Medical Images

Authors: Qusay Kanaan Kadhim1,A, Hamid sadeq Mahdi Alsultani2,A, Ihsan Salman3,A, Shaymaa Taha Ahmed4,A, Rasha Mahdi Abdulkader5,A

Department of Computer Science, University of Diyala, Baqubah 32001, Iraq

1 ORCID: 0000-0003-2814-2409, dr.qusay.kanaan@uodiyala.edu.iq

2 ORCID: 0000-0002-3571-9619, hamedsultani@uodiyala.edu.iq

3 ORCID: 0000-0002-0974-4271, drihsan@uodiyala.edu.iq

4 ORCID: 0000-0002-4986-2475, shaimaaAhmed@uodiyala.edu.iq

5 ORCID: 0009-0008-1658-1600, rashamahdi@uodiyala.edu.iq

 

Abstract

Monkeypox is a frequent viral disease difficult for diagnosis, as in medical images it has strong visual similarity with other skin diseases. Self-diagnosis can bring the high risks of mistakes, so employing artificial intelligence in medical image analysis could be a kind of helpful way to make quick and correct diagnosis. In this study, we present a new deep learning model for the diagnosis of monkeypox from imaging skin lesions. This system is a fusion of several active and content zones that incorporate CNN characteristics, attention mechanisms, and a SVM classifier. Use of multiple imaging scopes in medical image aids to capture variations in lesions at different angles, different tissue and illumination settings for extended diagnostic information. The CNN feature concatenation merges deep features from ResNet50 with light features of EfficientNetB0 to present a full view of pest imaging characteristics. To enhance the interpretability of the model, attention-based mechanisms such as the Convolutional Block Attention Module (CBAM) are applied to focus on diagnostically important portions of the lesion while reducing irrelevant background noise. Moreover, the standard SoftMax output layer can be replaced by a Support Vector Machine (SVM) classifier to provide more separated decision boundaries and increase the generalisation capability of the model. The framework we introduced achieved outstanding results on two medical imaging datasets with an overall accuracy of 98% superior to standard single CNN method. The primary contribution of this work is the presentation of a novel deep learning framework (hybrid Deep Learning) incorporating multi-view modalities, attention-based CNN feature fusion and strong SVM classification which address the limitations in state-of-the-art methods with respect to robustness, interpretability and clinical confidence.

 

Keywords: Monkeypox, Multi-View Images, Deep Learning, Convolutional Block Attention Module (CBAM), Support Vector Machine (SVM), Convolutional Neural Networks (CNNs).