Mpox Detection System using Deep Learning

A Web-Based Mpox Detection System Using Deep Learning. Sodiya Adesina Simon Department of Computer Science Federal University of Agriculture Abeokuta, Nigeria [email protected] Falana Olorunjube James Department of Computer Science. Federal University of Agriculture Abeokuta, Nigeria [email protected] Egwu Samuel Onuche-Ojo Department of Computer Science Federal University of Agriculture Abeokuta, Nigeria Imhonophi Gloria B. Federal Medical Centre Abeokuta, Nigeria [email protected] Awodele Oludele Babcock University Ilishan Remo Ogun State, Nigeria Aborisade Dada O. Department of Computer Science Federal University of Agriculture Abeokuta, Nigeria [email protected] Abdullahi Ahmad Federal Medical Centre Abeokuta, Nigeria, [email protected] Abstract The resurgence of monkeypox (mpox) has become a significant global health concern, underscoring the need for fast and reliable diagnostic tools. This study presents a deep learning system for automatic classification of skin lesions to identify mpox. A publicly available dataset containing four categories—Monkeypox, Chickenpox, Measles, and Normal— was cleaned and resized to 224 × 224 pixels for optimal MobileNetV2 performance. Using transfer learning, the original convolutional layers were frozen while custom classification layers were added. The model achieved an accuracy of 82.9%, with strong precision in distinguishing Monkeypox and Normal lesions. Weighted averages of precision (82.7%), recall (82.9%), and F1-score (82.8%) demonstrate balanced performance across all classes. A confusion matrix further supports the reliability of the results. To enhance usability, a Flask-based web interface was developed, enabling users to upload images and receive realtime predictions with confidence scores. The findings show that lightweight deep learning models can effectively support mpox identification, particularly in low-resource settings where accessible and efficient diagnostic methods are critical. Keywords— Monkeypox, Mpox, Artificial Intelligence, Deep Learning, MobilenetV2 I. INTRODUCTION Mpox (formerly known as monkeypox) is a viral disease caused by the Mpox virus, which belongs to the same family as smallpox. It mainly attacks animals but can also be transferred to humans by direct contact with infected animals or from human to human transmission. Historically known to occur in specific areas in Central and West Africa, Mpox has become a global health concern lately and has spread to non-endemic regions [1]. This makes timely and effective detection methods of great importance. Traditional methods of diagnosing the disease, including clinical evaluation and polymerase chain reaction (PCR) testing, while accurate, are time-consuming and require specialized equipment. These methods are often not within reach, particularly in places with poor healthcare infrastructure, and a delay in diagnosis can worsen the spread of the virus and create additional public health problems [2]. Artificial intelligence (AI), and deep learning in particular, has become a promising tool in the field of medical imaging. Convolutional Neural Networks (CNNs) have shown significant success in the automation of medical image detection and classification making them a suitable candidate for the diagnosis of dermatological conditions. By learning complicated patterns, such as edges, textures, and shapes, CNNs find them very useful for specific tasks, such as in the identification of diseased states [3]. While there has been successful application of deep learning techniques in recognition of skin cancers, there is a significant gap in research for detection of Mpox lesions, which have specific visual characteristics such as erythematous rashes, pimples and scab formation that occur in a specific temporal order [3]. This study seeks to address that gap in research by developing a web-based system to detect Mpox using deep learning algorithms to analyze images of skin lesions. Through the use of CNNs, the proposed system will support a more efficient and accurate diagnostic process that can promote more feasible medical intervention at the appropriate time. Moreover, with a web-based platform, the system will be available for healthcare professionals even in areas with limited resources. This accessibility will help to improve the rate of diagnosis, especially in areas that are remote and traditional diagnostic tools may not be readily available or accessible. The ultimate goal is to improve early detection of Mpox and offer a scalable, user-friendly solution which can be used globally and in countries with limited healthcare infrastructure in particular. The study contributes to existing knowledge in several key ways: Development of a Specialized Deep Learning Model: We have developed a deep learning model specifically designed for detecting Monkeypox (Mpox) lesions, leveraging the distinct visual features of the disease. The model is based on MobileNetV2 and transfer learning and can classify skin lesions with high accuracy among Monkeypox, Chickenpox, Measles and Normal. 2 Introduction of a Web-Based Platform: Monkeypox has been made easier to detect now with the introduction of a web-based platform. The service allows users to upload pictures and receive predictions immediately. It has a user-friendly interface that can operate in a low-resource environment so far as it has been found to identify Monkeypox early. Scalable Solution for Low-Resource Areas: Our model provides a unique scalable and easy-to-use medium, which can be implemented in areas with weak healthcare facilities. Using AI and a web-based application, more people could be exposed to monkeypox, so the early detection can be more effective. Foundation for Future AI-Based Disease Detection Systems: The research techniques that were reached in this study can be applied to detect other viral skin diseases and might be integrated into global surveillance initiatives on health. It can be used as a support mechanism to conduct real-time monitoring and act quickly in the event of future outbreaks like the pandemic. The outline of the paper is as follows. Section 2 provides a literature review, which includes the existing research on the diagnosis of Mpox and the use of artificial intelligence in medical imaging. Section 3 provides the description of methods, including the system design, data collection, preprocessing procedures, and the development of the deep-learning model. Section 4 discusses the result. Section 5 provides the conclusion and future directions. II. REVIEW OF RELATED WORKS The current outbreak of monkeypox (Mpox) has become a major public health concern due to its rapid spread across multiple countries. Early detection and diagnosis of Mpox are critical for effective treatment and management. Zhang et al. (2025) compared various pre-trained deep-learning architectures: VGG19, VGG16, ResNet50, MobileNetV2 and EfficientNetB3 on Mpox detection. The findings indicated that MobileNetV2 gave the finest results producing 98.16 per cent accuracy and abnormally high values in recall, precision, and F1-score indices [4]. Yasmin et al. (2024) proposed a more recent study that proposed a transfer-learning-based PoxNet22, which is a deeplearning-based classifier of Mpox lesions. After pre-training a data-augmentation set of data, the researchers narrowed a list of pre-trained models, such as VGG19, VGG16, ResNet50, MobileNetV2, EfficientNetB3, and EfficientNetV2, to minimize overfitting, achieving 100% precision, recall, and accuracy in classifying Mpox lesions [5]. Kundu et al. (2024) suggested a federated -learning model that is based on deep-learning models to identify Mpox and other pox viruses and protect the privacy of patients. It applies a Cycle-Consistent Generative Adversarial Network (CycleGAN) to augment the data by adding artificial samples of Mpox lesions to the training set. MobileNetV2, Vision Transformer (ViT), and ResNet50 were tested in this environment, with the ViT-B32-based demonstrating an impressive 97.90% accuracy, demonstrating the effectiveness of federated learning to make secure and accurate detection possible [6]. The authors used MobileNetV2, Vision Transformer (ViT) and ResNet50 for the classification tasks, taking advantage of how well these models have been shown to work in image recognition. These models were thereafter trained and evaluated in a federated learning framework, in which several models are trained collectively across decentralized systems without requiring the exchange of raw data, which ensures privacy and security and highlights the prospects of incorporating federated learning with deep neural networks for the secure and accurate detection of Mpox [6]. Nayak et al. (2024) benchmarked five pre-trained convolutional neural networks (i.e. GoogLeNet, Places365-GoogLeNet, SqueezeNet, AlexNet, and ResNet-18). Hyperparameter optimisation showed that ResNet - 18 gave the best accuracy signalled by 99.49 percent, all the evaluated models had verification accuracies of more than 95 percent. The study also used techniques such as LIME and Grad-CAM which are explainable AI methods that would help clinicians to interpret the rationale behind the predictions of each model [7]. Alghoraibi et al. (2025) presented an AI-based Mpox identification on a skin image detection system, called ITMAiNn. Its architecture consists of an AI pipeline that uses transfer learning to use on public datasets, a cross-platform mobile application, and a real-time public-health dashboard. In binary classification (Mpox vs. non-Mpox), Vision, MobileViT, Transformer in Transformer, and VGG16 achieved an accuracy of 97.85%. Multiclass detection (Mpox, cowpox, measles, healthy skin, chickenpox, hand-foot-disease) ResNetViT and ViT Hybrid detected with 92 % accuracy [8]. The MobileViT model is also implemented on the mobile application and enables the user to scan lesions, monitor the symptoms, and find the closest healthcare facility through GPS. Combined with the real-time dashboard, health authorities may track the development of cases and coordinate interventions, showing how deep learning, mobile technology, and analytics will empower the infrastructure of a smart city in terms of population health. In Taspinar et al. (2024), Mpox skin-lesion detection was conducted on VGG16 and VGG19 CNNs. VGG19 (97.81 percent) was better than VGG16. The study revealed that transfer learning and fine-tuning have a significant positive effect on performance. Grad-CAM visualisation also helped determine what image regions affected the choices of the model's purposes further [9]. Gupta et al. (2024) developed an Mpox detection system that used GoogLeNet, EfficientNet, and VGG19. An open-source collection of skin-lesion images was used, which went through transfer learning and data augmentation; tenfold crossvalidation was done afterward, and it was seen that the models 3 were similar in their performance. EfficientNet became the most efficient architecture and achieved high accuracy, sensitivity, and specificity, which justifies the credibility of convolutional neural networks in the task of Mpox image recognition. [10]. Thieme et al. (2024) employed a convolutional neural network named MPXV-CNN, where the convolutional neural network is designed for the early diagnosis of Mpox virus (MPXV) lesions. The model has been taught using a large dataset of 139,198 images of skin lesions. MPXV-positive images (676 in total) were pulled from various different sources: peerreviewed literature, media reports, social media platforms and a prospective cohort from the Stanford University Medical Center (63 images from 12 male patients). The remaining 138,522 images of non-MPV lesions were taken from 8 dermatological repositories [11]. The data set was divided into the four data sets: train data, validation data, and test data, which would show the performance of the model in the best possible manner [11]. MPXV-CNN showed excellent diagnostic performance in the validation and testing dataset with a sensitivity of 0.83 and 0.91, a specificity of 0.965 and 0.898, and an area under the receiver operating characteristic curve (AUC) of 0.967 and 0.966, respectively. In the prospective cohort, a sensitivity of 0.89 was reached by the model. These outcomes validate the fact that MPXV-CNN achieves adequate findings in Mpox lesion identification despite a heterogeneous skin tone and anatomical regional spectrum [11]. Chakroborty (2024) made changes by proposing a hybrid deep learning system with an image feature extractor and a traditional machine learning classifier to use representation learning as well as classical predictive modeling systems. Deep features were generated using MobileNetV2, and final classification was provided in a LightGBM model. This learning architecture had an accuracy of 91.49, a F1-score of 90.91, weighted precision of 91.87, a weighted recall of 91.49, and a Matthews correlation coefficient of 0.83, indicating the usefulness of hybrid models to the early detection of Mpox [12]. Onyema et al. (2024) also included the issue of hair artifacts in dermoscopic pictures. Their first publication, Mpox Skin Lesion Dataset (MSLD), included the images of Mpox, chickenpox, and measles. The enhanced U-Net was trained following hair-removal preprocessing that achieved 90% accuracy, 89% recall, and 86% F1-score on Mpox and outperforms the current models by an order of magnitude and allows proper diagnostics in difficult conditions [13]. Nath and Moazzam (2024) offered a different hybrid design to the model called MpoxNet that combines the images with tabular symptoms. It is a risk predictive system that uses LSTM networks and multilayer perceptrons to predict the likelihood of Mpox, particularly in HIV positive patients. Following class balance by resampling, MpoxNet reached 65.35, 87.50, and 100 percent accuracy in Dataset D1, D2, respectively, and 100 percent recall, which was superior to traditional examples like AdaBoost, XGBoost, and Random Forest [14]. The gap in the literature clearly illustrates the need for a webbased Mpox detection system that combines high classification accuracy, robust segmentation, interpretability, and real-world deployment. Such a system would not only address the limitations of current models but also provide a valuable tool for early diagnosis and control during outbreaks. III. METHODS 1. Dataset Description The Monkeypox Skin Image Dataset was designed to help identify Monkeypox (Mpox) early on by classifying the skin lesions in four categories: Monkeypox, Chickenpox, Measles, and Normal. The Department of Computer Science and Engineering at Islamic University, Kushtia, Bangladesh compiled the dataset using pictures obtained through various sources over the internet such as medical picture libraries, academic literature as well as other web repositories. [15]. The data set consists of annotated images of every category, each picture has been rescaled to sizes of 224 X 224 pixels to fit the input requirements of the employed deep-learning framework, and in the training phase a sample population of 152 images, distributed across four categories: Monkeypox (55 images), Chickenpox (21 images), Measles (18 images), and Normal (58 images). The dataset was later divided into training, evaluating and test sets after an 80-20 split algorithm, eighty percent of the observations contained in the training group and twenty percent in the evaluating group, to mitigate over-fitting and awareness of the model [15]. 2. Model Architecture We have used MobileNetV2 as the base architecture, where we have adopted this lightweight and computationally efficient convolutional neural network in this project of detecting Monkeypox. The network was trained with a set of weights that are pre-trained on ImageNet and later fine-tuned through transfer learning to be used in the particulars of Mpox images. To assist with the targeted classification requirements, layers from the top of the classification were removed, thereby opening the possibility of removing the specially engineered layers developed for the detection of Mpox. The architecture includes the following: Base Model: The pre-training MobileNetV2 architecture with top layers of classification removed (include_top= True) was used as a base model which takes advantage of the feature representations learned during ImageNet pre-training while giving a basis for adding customized classification layers for future training. Freezing the Base Model: In order to retain the discriminative features learnt from ImageNet, the weights of the MobileNetV2 4 layers were frozen, thus making them immutable during the current training procedure. Custom Classification Layers: Global Average Pooling: This is a technique that diminishes the spatial dimension of the feature maps, keeping the salient points, so it reduces the number of parameters, which also avoids the problem of overfitting. Dense Layer (1024 Units): A complete-layer of 1,024 units with Rectified Linear Unit (ReLU) was added to model complicated, non-linear features that are found in the data. Dropout (0.5): Dropout regulariser with a keep-probability of 0.5 was used, causing a stochastic process by which 50% of the neurons will be switched off during one iteration of training to reduce the problem of over-parameterisation. Softmax Output Layer: The final classification layer was a dense layer with four out layers (Monkeypox, Chickenpox, Measles, and Normal) with the softmax activation function to get class-specific probability distributions. Training was done using the Adam optimiser along with categorical cross-entropy as the loss function, a common choice for multiple class classification problems. The proposed system follows a sequence of steps, from image upload to prediction and result display. A flowchart of the system is shown below: Fig 1: Flowchart of the Proposed System. 3. Model Training Using the training dataset, the model was then trained for 10 epochs. A batch size of thirty-two has been used to achieve a tradeoff between computational efficiency and model performance. There were the following important components to the training process: Adaptation: The Adam optimizer has been chosen because it has an adaptive learning rate that makes it more effective for training deep learning models. Fig 2: Learning Rate vs. Epoch This figure shows the learning rate used during model training over 10 epochs. In this experiment, we used a constant learning rate of 0.001 throughout the training process. The learning rate controls the size of the steps the model takes when adjusting weights during training. A constant learning rate ensures stable model updates and avoids drastic weight changes. Loss Function: Categorical cross entropy loss function is the aspect that has been utilized with appropriate consideration of this multi-class classification problem. Metrics: Accuracy was used to evaluate the model because it can be used to track the model's performance both during training and for validation. Enhancing verification quality: The above techniques and others for augmenting data such as rotation, zoom, horizontal flipping the augmentation process were applied to increase the variety of the training dataset and hence the model will have a better ability to generalize. The model's performance was monitored on a validation set during training coerce to prevail overfitting. 4. Model Evaluation Validation data were used to assess the performance of the model using a suite of performance metrics such as accuracy, precision and recall, F1-score, and the confusion matrix. These metrics were calculated for each class which include Monkey Pox, Chicken Pox, Measles and Normal separately and in percentage. 5 For the former, it is important to consider the following: • Accuracy: The percentage of correct cases that were correctly classified. • Precision: The fraction of actual positive results welcomed from the list of predicted positive results. • Recall: Fraction of true positive out of all actual positive that are correctly classified • F1-Score: A balance between precision and recall which is the harmonic mean of the two values i.e., achieves harmonious analysis • Confusion Matrix: Used to calculate true positive, false positive, true negative and false negative for each class. Beyond evaluation on the validation set, classification reports and confusion matrices were generated in order to show how the model performed among the four classes. 5. Model Deployment Following the model training and evaluation, the model was implemented as a web application in Flask. The application allows users to scan the skin lesions and gets model predictions in real-time. Users are shown the class to which they are predicted to belong - either Monkeypox, Chickenpox, Measles or Normal - and the confidence level as a percentage. The trained model, its architecture and the labels of classes were saved in three different formats: • mpox_model.h5: Contains the trained model and its weights. • mpox_model.json: Contains the model architecture. • labels.txt: Contains the class labels. These files were integrated into the Flask web app, enabling users to interact with the model and obtain predictions directly through a user-friendly interface. IV. RESULTS 1. Model Performance The performance of the model was evaluated using some quantitative metrics, such as accuracy, precision, recall, F1score, and confusion matrix. The overall accuracy rate of the model was 82.9 % which proved the use of this model in distinguishing four diagnostic classes: Monkeypox, Chickenpox, Measles, and normal. Fig 3: Training and Validation Accuracy of the Proposed Model. This figure demonstrates how the model's accuracy increased over the course of 10 epochs, with a final training accuracy of 82.9% and a validation accuracy of 79%. Fig 4: Training and Validation Loss of the Proposed Model. This figure illustrates the decrease in loss for both the training and validation datasets, suggesting that the model is learning effectively without overfitting. Table 1: Model Performance Summary. Class Precision Recall F1Score Support Chickenpox 0.684 0.619 0.650 21 Measles 0.737 0.778 0.757 18 Monkeypox 0.818 0.818 0.818 55 Normal 0.915 0.931 0.923 58 6 Macro Avg 0.789 0.787 0.787 152 Weighted Avg 0.827 0.829 0.828 152 2. Confusion Matrix The confusion matrix for the model’s predictions is shown below. The matrix highlights how well the model distinguishes between Monkeypox, Chickenpox, Measles, and Normal. Fig 5: Confusion Matrix of the Proposed Model. 3. Distribution of Prediction Probabilities To improve our understanding of the model's prediction outcomes, we investigated the distribution of the predictive probabilities between all classes. The figure below shows the probability distributions of predictions of the four classes. Fig 6: Distribution of Prediction Probabilities of the proposed model. Figure 6 shows that Monkeypox and Normal classes show different peaks, meaning high confidence of predictions in these categories. Chickenpox and Measles show some more overlap of prediction probabilities which probably helps in misclassification that is sometimes happening in the confusion matrix. 4. Multi-Class ROC Curve We also evaluated the Receiver Operating Characteristic (ROC) curve of each class to better understand the model's ability to differentiate between classes. The multi-ROC curve shown below shows the True Positive Rate (TPR) as a function of the False Positive Rate (FPR) at each class. Fig 7: ROC Curve of the proposed model. Figure 7 shows that Normal class has the highest AUC (Area Under the Curve) at 0.98, indicating that the model is highly confident in distinguishing Normal from other classes. Monkeypox and Measles show good performance with AUCs of 0.94 and 0.95, respectively. Chickenpox also performed well, with an AUC of 0.89. 5. Model Limitations and Observations Although the model performed well in general, there were several areas for improvement noted. Class Confusion: The distributions of the prediction probabilities of the Measles and Chickenpox classes overlapped, which resulted in some misclassifications. Minor Class Misclassifications: Although the model is robust, minor misclassifications in differentiating between similar diseases (e.g. between Measles and Chickenpox) could be reduced by considering and exploiting more features such as data from the patient's age or symptoms, or by augmenting the data used for training with representative images. 7 The model was then converted into a web application for sending images of skin lesions so that they can be predicted. 6. Performance Benchmarks The following performance benchmarks were recorded for the model, highlighting its efficiency and suitability for deployment in low-resource environments: Model Latency: The model takes approximately 40 milliseconds to generate a prediction for a single image. This ensures a fast response time, making the model suitable for applications requiring quick decision-making. Model Size: The total size of the trained model is 15MB, which is relatively small. This compact size makes it ideal for deployment in lowresource environments, where computational and storage resources may be limited. Training Time: The model was trained for 10 epochs, with a total training time of 5 minutes (300 seconds). This rapid training time demonstrates the model's computational efficiency while still achieving high accuracy and generalization. Inference Time: The model can make predictions in approximately 20 milliseconds per image, which is fast enough to support realtime applications in environments such as mobile devices or embedded systems. Overall, the model had relatively good performance (accuracy of 82.9%), good precision and recall for the Monkeypox and normal classes. The confusion matrix, prediction probability distributions and the ROC curves combined provide an overall analysis of the performance of the model and pinpoint possible areas for improvement, in particular, distinguishing Chickenpox from Measles. V. CONCLUSION & FUTURE DIRECTIONS This research explores the development of an early detection of Monkeypox (Mpox) with help of skin lesions using deep learning-based images. The system is based on a MobileNetV2 architecture, that was trained on ImageNet and then fine-tuned for the multi-class classification problem of Monkeypox, Chickenpox, Measles and Normal categories. Transfer-learning techniques were used, using the pretrained weights of the MobileNetV2, which significantly improved the performance of the model, even with the relatively limited size of the dataset. The model achieved 82.9% accuracy. Analysis of confusion matrix shows that the misclassification has occurred mostly between Measles and Chickenpox; however, these observations can be overcome by increasing variety and balance of training data. The trained model was successfully integrated to the web application, and it allows the healthcare professionals to upload skin lesion images and avail real time model predictions with confidence scores. This tool provides a promising avenue of assisting in the early detection of Monkeypox and thus, early interventions and containment of the disease. The proposed system has great promises in clinical applications, where proper and fast diagnosed results of Monkeypox can play a role in public health surveillance and disease control measures. Moreover, the model implementation via web-based interface provides the accessibility for medical professionals as well as its deployment in resource constrained devices, including smartphones. Future research efforts could be focused on expanding the data set to include more skin pigmentation and more pathological manifestations to improve generalizability. Additionally, the inclusion of clinical datasets including patient symptomatology and historical medical records could also give rise to opportunities to improve the accuracy of predictive models and to provide support for more complex and sophisticated prognostications [16]. The system may be also extended using real-time updating functionality and supporting the mobile applications, and so medical personnel working in a remote environment can use the imaging data transmission for evaluation. Furthermore, the localising of lesions through multi-class segmentation techniques is expected to strengthen the diagnostic precision to enable it to be a more complete diagnostic tool. Finally, future investigations should include clinical validation protocols and compliance with regulatory certification procedures to ensure the reliability and security of system when operating in the real-world healthcare setting. [17]. ACKNOWLEDGEMENT This is a research work of Artificial Intelligence for One Health (AIA4OneHealth), also called AI4PEP Nigeria. This work has been funded by the Canadian International Development Research Centre (IDRC) under the Grant Agreement 109981, funded by IDRC and the UK government’s Foreign, Commonwealth and Development Office. The views expressed herein do not necessarily reflect those of the UK government’s Foreign, Commonwealth and Development Office, IDRC or IDRC’s Board of Governors. REFERENCES [1] “Mpox.” [Online]. Available: https://www.who.int/news-room/factsheets/detail/mpox [2] N. N. Soe et al., “Using AI to Differentiate Mpox From Common Skin Lesions in a Sexual Health Clinic : Algorithm Development and Validation Study Corresponding Author :,” vol. 26, pp. 1–12, 2024, doi: 10.2196/52490. [3] S. Vuran, M. Ucan, M. Akin, and M. Kaya, “Multi- 8 Classification of Skin Lesion Images Including Mpox Disease Using Transformer-Based Deep Learning Architectures,” 2025. [4] A. S. Jaradat et al., “Automated Monkeypox Skin Lesion Detection Using Deep Learning and Transfer Learning Techniques,” no. January 2019, 2023. [5] F. Yasmin et al., “PoxNet22 : A Fine-Tuned Model for the Classification of Monkeypox Disease Using Transfer Learning,” IEEE Access, vol. 11, no. March, pp. 24053–24076, 2023, doi: 10.1109/ACCESS.2023.3253868. [6] D. Kundu, M. Rahman, A. Rahman, and D. Das, “Federated Deep Learning for Monkeypox Disease Detection on GAN-Augmented Dataset,” IEEE Access, vol. PP, p. 1, 2024, doi: 10.1109/ACCESS.2024.3370838. [7] Nayak, K. Chadaga, N. Sampathila, H. Mayrose, N. Gokulkrishnan, M. B. Ga, S. Prabhu, S. K. S., and S. Umakanth, "Deep learning based detection of monkeypox virus using skin lesion images," Medicine in Novel Technology and Devices, vol. 18, p. 100243, Jun.2023. https://doi.org/10.1016/j.mntd.2023.100243. [8] H. Alghoraibi, N. Alqurashi, S. Alotaibi, R. Alkhudaydi, B. Aldajani, and J. Batawil, “Deep Learning-Based Mpox Skin Lesion Detection and RealTime Monitoring in a Smart Healthcare System †,” no. September 2022, pp. 1–26, 2025. [9] Y. S. Taspinar, I. Cinar, R. Kursun, and M. Koklu, “Monkeypox Skin Lesion Detection with Deep Learning Models and Development of Its Mobile Application,” vol. 12, no. 1, pp. 273–285, 2024. [10] M. Gupta and R. Tiwari, “e-Prime - Advances in Electrical Engineering , Electronics and Energy Automated detection of human monkeypox using skin lesion images and transfer learning CNN,” e-Prime - Adv. Electr. Eng. Electron. Energy, vol. 14, no. March 2024, p. 101113, 2025, doi: 10.1016/j.prime.2025.101113. [11] A. H. Thieme et al., “A deep-learning algorithm to classify skin lesions from mpox virus infection,” vol. 29, no. March, 2023, doi: 10.1038/s41591-023-022257. [12] S. Chakroborty, “H-MpoxNet : A Hybrid Deep Learning Framework for Mpox Detection from Image Data,” no. Ml, 2025. [13] E. M. Onyema et al., “Deep learning model for hair artifact removal and Mpox skin lesion analysis and detection,” pp. 1–18, 2025. [14] T. D. Nath and G. Moazzam, “International Journal of Cognitive Computing in Engineering MpoxNet : Leveraging hybrid deep learning for enhanced monkeypox diagnosis and risk identification,” Int. J. Cogn. Comput. Eng., vol. 7, no. 1, pp. 86–94, 2026, doi: 10.1016/j.ijcce.2025.09.001. [15] D. Bala, S. Hossain, and M. Alamgir, “MonkeyNet : A robust deep convolutional neural network for monkeypox disease detection and classification,” Neural Networks, vol. 161, pp. 757–775, 2023, doi: 10.1016/j.neunet.2023.02.022. [16] Y. Gulzar, S. Agarwal, and S. Soomro, “Nextgeneration approach to skin disorder prediction employing hybrid deep transfer learning”. [17] W. Hong, “Advances and Opportunities of Mobile Health in the Postpandemic Era : Smartphonization of Wearable Devices and Wearable Deviceization of Smartphones Recent Progress in the Development of Wearable Electronics for Health,” vol. 12, pp. 1–22, 2024, doi: 10.2196/48803.