Monkeypox Detection and Classification Using Deep Learning Based Features Selection and Fusion Approach

In today's healthcare system, clinical diagnosis has taken on a crucial role. As the COVID-19 virus's global infection declines, the monkeypox virus is steadily developing. Because of this, it's critical to identify them early, before they spread to the larger population. Early detection can be aided by AI-based detection. In this study, a fusion based contrast enhancement approach is used to preprocess the source images. Two pre-trained DCNN models (Inception-ResNet-V2 and NASNet-Large) are modified and trained using transfer learning. From each DCNN model, deep feature vectors are extracted and the entropy-based controlled algorithm is used for the best features selection. The convolutional sparse image decomposition fusion approach is utilized to fused the feature for classification. Finally, the selected features are forwarded to a multi-class support vector machine (M-SVM) for final classification. After performing experiments on public datasets, the proposed approach obtained an accuracy of 98.59%, sensitivity of 92.78%, specificity of 95.47%, and AUC of 0.987. Simulation studies show that the proposed approach outperforms other methods both visually and quantitatively.