Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Scientific Reports volume 15, Article number: 1746 (2025)
Metrics details
Skin diseases impact millions of people around the world and pose a severe risk to public health. These diseases have a wide range of effects on the skin’s structure, functionality, and appearance. Identifying and predicting skin diseases are laborious processes that require a complete physical examination, a review of the patient’s medical history, and proper laboratory diagnostic testing. Additionally, it necessitates a significant number of histological and clinical characteristics for examination and subsequent treatment. As a disease’s complexity and quantity of features grow, identifying and predicting it becomes more challenging. This research proposes a deep learning (DL) model utilizing transfer learning (TL) to quickly identify skin diseases like chickenpox, measles, and monkeypox. A pre-trained VGG16 is used for transfer learning. The VGG16 can identify and predict diseases more quickly by learning symptom patterns. Images of the skin from the four classes of chickenpox, measles, monkeypox, and normal are included in the dataset. The dataset is separated into training and testing. The experimental results performed on the dataset demonstrate that the VGG16 model can identify and predict skin diseases with 93.29% testing accuracy. However, the VGG16 model does not explain why and how the system operates because deep learning models are black boxes. Deep learning models’ opacity stands in the way of their widespread application in the healthcare sector. In order to make this a valuable system for the health sector, this article employs layer-wise relevance propagation (LRP) to determine the relevance scores of each input. The identified symptoms provide valuable insights that could support timely diagnosis and treatment decisions for skin diseases.
A severe worldwide health concern that affects a large number of people is skin disease. Due to their physical and psychological impacts on individuals, skin diseases are a serious and concerning problem in societies1. Early-level detection of the kind of disease is vital in determining the appropriate treatments that can be carried out.
Varicella zoster virus (VZV) causes chickenpox2. VZV belongs to the herpes virus family3. The majority of affected children are between 6 months and 7 years old4. Chickenpox, the highly contagious disease that affects children, is now being reported to affect significant portions of adults5. The main symptom is an itchy, red rash that frequently turns into blisters and progresses to flu-like symptoms. It spreads through contact with sick individuals, contaminated objects, and airborne droplets brought on by coughing or sneezing. Additionally, chickenpox can be dangerous and even fatal, particularly in pregnant women6.
High fever, cough, coryza, conjunctivitis, and a morbilliform rash are all symptoms of the highly contagious and possibly fatal airborne measles7. Measles is a fatal viral disease carried on by a paramyxoviridae family RNA virus8. Despite being seen as a childhood disease, measles can affect people of all ages. A single measles case has been shown to result in 12–18 secondary infections in an otherwise healthy group9.
With the coronavirus disease (COVID-19) pandemic still raging, just as people are beginning to adjust to the ‘new normal,’ the monkeypox virus is wreaking havoc on the world. Healthcare professionals around the globe are progressively concerned regarding the most recent monkeypox outbreak. The Democratic Republic of the Congo (previously Zaire) received the first human monkeypox infection report in 197010. The monkeypox virus is a zoonotic pathogen that renders a rash akin to that of smallpox11. According to research, the Poxviridae family, of which the monkeypox virus is a member, was first spread from animals to people12. It can be spread by respiratory droplets, animal bites, mouth, nose, or eye mucus, or by direct contact with another person13.
Compared to COVID-19, monkeypox is not as contagious, although the number of cases is still rising. In 1990, there were just 50 cases of monkeypox recorded in West and Central Africa14. However, 5,000 cases were reported in 2020. Despite the consensus that monkeypox only happened in Africa, cases of the virus were documented in 2022 in several non-African countries in Europe and the US15. Because of this, people are progressively growing more fearful and nervous, which typically shows in their thoughts on social media. Scientists attribute the current outbreak of monkeypox in humans, thought to be occurring on a global scale, to either a change in the essential characteristics of the monkeypox virus or adjustments in the human lifestyle16.
Lately, there has been a significant growth in the amount and quality of research in many areas using machine learning (ML) to classify skin lesions17. However, to improve ML models, effective feature extraction techniques are required. Traditional ML classifiers have the disadvantage that one must create complex hypotheses independently, whereas deep neural networks generate them automatically, making them an effective tool for learning non-linear correlations18. Due to their increased success in processing enormous amounts of data and the capability to extract hidden valuable knowledge from data, deep learning (DL) approaches have historically diverged from traditional ML techniques19. The DL has been successfully utilized in numerous areas, like lesion detection20,21,22, classification23,24,25,26,27,28,29, and segmentation of medical images30,31.
Additionally, hybrid approaches such as Adaptive Neuro-Fuzzy Inference Systems (ANFIS) have been applied to tasks like sentiment analysis32. Convolutional neural networks (CNNs) are generally utilized in DL to learn features automatically and use that information for classification. CNNs are a class of neural networks typically used on data where the input has another tensor structure, such as an image. They are specifically designed to capture the inherent structure of images and visual data through a sequence of Modules with interconnected Nodes, which automatically generate spatial hierarchies of features, making them ideal for image recognition, image segmentation, object detection, and image classification problems.
Skin images of four classes: (a) Chickenpox; (b) Measles; (c) Monkeypox; (d) Normal.
Given that DL is the state-of-the-art for analyzing medical images33,34, it is not surprising that medical professionals have expressed their concerns regarding the technology’s “black box”35. The need for more transparency and fairness in DL models is a notable issue, specifically in critical areas such as healthcare, where trust and understanding of artificial intelligence (AI) predictions are vital. To tackle this issue, explainable artificial intelligence (XAI) has emerged as a powerful approach, allowing AI models to produce results that people can understand and trust. This article proposes a model for identifying and predicting skin diseases using transfer learning (TL) empowered with XAI. The primary objective is to address the shortcomings of current methods, such as low accuracy and a lack of explainability, by creating a model that is more precise and easier to interpret. The expected result is a more accurate and interpretable AI model, addressing the common issues of low performance and lack of transparency in current healthcare applications. TL employs pre-trained CNNs, utilizing learned features to facilitate the adaptation of models to new tasks, hence minimizing the training time and data needs. XAI refers to AI approaches that produce results humans can interpret or comprehend. Figure 1 shows chickenpox, measles, monkeypox, and normal skin images.
The remainder of the article is divided as follows: Sect. “Literature review” discusses the literature review, Sect. “Materials and methods” discusses materials and methods, Sect. “Simulation and results” offers the simulation and results, Sect. “Practical and managerial implications” gives practical and managerial implications, and finally, Sect. “Conclusion and future work” provides the conclusion and future work.
Skin disease is one of the most widespread diseases among people. Skin diseases range from superficial acne to severe conditions like squamous cell carcinoma. It affects people of all cultures, regions, and age groups. In the last decade, skin and subcutaneous diseases have been the fourth leading reason for the worldwide burden of non-fatal diseases36. Despite involving most of the population at a time, they do not get much importance from a public health point of view. The Global Burden of Disease Study 2017 reported that years of life lost due to skin disease are between 30 and 40 months for an adult37. Recognizing disease is crucial for choosing the most appropriate treatment and preventing its spread.
The pertinent data on the age and gender of chickenpox patients reveals that there is no real difference in the gender of individuals who are affected. Children, however, are the age group with the highest prevalence of chickenpox, mainly because they exhibit group social features and are more likely to spread the disease to those around them. Currently, vaccination is the only method of mass prevention that has been scientifically shown to be both practical and economical. In the US, a varicella vaccine surveillance study found that 13% of children between the ages of 5 and 10 are susceptible to the disease. Similarly, studies conducted in the UK show that approximately 40% of kids aged 1 to 9 are vulnerable to infection, but less than 10% of kids over 15 are38. The disease burden has significantly decreased in developed nations, and most developed nations have included the varicella vaccine as part of their standard immunization plan39. Roy et al.40 used various segmentation approaches to identify skin diseases like chickenpox, candidiasis, cellulitis, and acne.
A severe, contagious viral disease is measles. Before the measles vaccine was created in 1963, massive measles epidemics happened every two to three years, resulting in an estimated 2.6 million measles fatalities yearly. However, between 2000 and 2016, there was an 84% decline in measles mortality as the vaccine became more readily available41. Vaccination is a reasonably effective way to avoid measles.
Until 1958, reports of smallpox-like diseases in monkeys were sporadic, and monkeypox was relatively unknown42. Intense smallpox surveillance in the Democratic Republic of the Congo, where smallpox was considered eliminated, led to the identification of the first human case of monkeypox in 1970. The patient, a nine-month-old boy with hemorrhagic monkeypox, survived the infection43. Before 2003, only African countries had recorded human cases of monkeypox44. A multi-state zoonotic outbreak in the USA that lasted from May to June 2003 was the cause of the first human cases of monkeypox outside of Africa45. Monkeypox outbreaks have been documented in several nations, primarily in Europe, since the beginning of May 2022, although the monkeypox virus is not prevalent in those regions46.
The ‘Monkeypox Skin Lesion Dataset (MSLD)’ was developed by Ali et al.47 and includes skin lesion images of chickenpox, measles, and monkeypox, with the majority of images sourced from publicly accessible case reports, blogs, and news websites. The sample size is expanded through data augmentation, and a 3-fold cross-validation experiment is set up. Different pre-trained DL models, including VGG16, ResNet50, and InceptionV3, are used to classify monkeypox and other diseases. Additionally, an ensemble of the three models is created. VGG16, ResNet50, InceptionV3, and ensemble achieved accuracies of 81.48 (± 6.87%), 82.96 (± 4.57%), 74.07 (± 3.78%), and 79.26(± 1.05%), respectively. Burak Gülmez48 developed a hybrid DL model, “MonkeypoxHybridNet,” by combining three pre-trained models—ResNet50, VGG19, and InceptionV3. This model was trained on the “Monkeypox2022” dataset and attained an accuracy of 84.2%.
Irmak et al.49 utilized pre-trained DL architectures to detect monkeypox skin lesions. This study’s classification used the monkeypox skin image dataset, which was open-sourced in 2022. The dataset contains four classes: chickenpox, measles, monkeypox, and normal. Pre-trained DL architectures, MobileNetV2, VGG16, and VGG19, were trained. MobileNetV2 had the best performance result, with an accuracy of 91.38% compared to VGG16 and VGG19.
Singh and Songare50 applied the DL models InceptionV3, GoogLeNet, ResNet50, and VGG16 to a two-class dataset containing normal and monkeypox classes and discovered that the GoogLeNet model had the highest accuracy at 88.27%. Sharma et al.51 developed a custom ResNet-18-based model for detecting monkeypox, measles, and chickenpox and compared it to several other models. Their model’s accuracy was 84.59%. Using Darknet 19 and Improved Darknet 19, Sethy et al.52 suggested a novel technique for the early diagnosis of monkeypox in their study. The research dataset included samples of skin diseases like chickenpox, measles, monkeypox, and normal cases. Darknet 19 and Improved Darknet 19 models were reported to have attained accuracies of 81.4% and 85.49%, respectively.
Uysal53 created a hybrid AI system capable of detecting monkeypox in skin images. This dataset contains four classes. In the original dataset, the data distribution of the classes is unbalanced. Several data augmentation and data preprocessing techniques were employed to rectify this disparity. The test accuracy of the hybrid AI system devised and suggested for monkeypox detection was 87%. Ariansyah et al.54 suggested a CNN and VGG16-based classification methodology to identify the symptoms of monkeypox and measles. The image dataset used in this proposed methodology contains the classes of monkeypox, measles, and normal. VGG16 achieves a high accuracy of 83.33% as compared to CNN.
Kundu et al.55 proposed an ML and DL classification methodology for monkeypox prediction. Monkeypox and others (which include chickenpox or measles) are two categories of skin lesions that are included in the dataset. Support vector machine (SVM) and k-nearest neighbor (KNN) were utilized as ML algorithms, while vision transformer (ViT) and RestNet50 were utilized as DL algorithms. Among the ML models, the KNN attains the best accuracy of 84%. However, with an accuracy of 93%, the ViT acts better than the other models.
Aqsa Akram et al.56 introduced “SkinMarkNet,” a novel technique for classifying monkeypox lesions utilizing an ensemble of three TL models—Inception, Xception, and ResNet. The study addresses the scarcity of annotated data by using data augmentation techniques, which enhance the training dataset and improve the model’s performance. The dataset, consisting of diverse skin lesion images from the Kaggle repository, was used to train the model. “SkinMarkNet” achieved a high classification accuracy of 90.615%, outperforming traditional ML and DL methods. The research shows the prospect of combining advanced DL models and data augmentation to enhance the automated diagnosis of monkeypox, contributing to more effective public health responses.
There are a few prominent limitations regarding the previous research, as given in Table 1.
There is area for improvement in the overall accuracy of previous literature47,48,49,50,51,52,53,54,55,56
No use of explainable artificial intelligence47,48,49,50,51,52,53,54,55,56
The noteworthy contributions of this proposed article are as follows:
Skin diseases have been identified and predicted using this proposed model.
The proposed model classifies chickenpox, measles, monkeypox, and normal skin images into their respective classes.
The performance metrics for the proposed model demonstrate encouraging outcomes, including accuracy, misclassification rate, precision, specificity, sensitivity, false negative rate (FNR), false positive rate (FPR), and F1 score.
The main contributions of this proposed model are improved accuracy relative to previous works and the incorporation of the XAI approach layer-wise relevance propagation (LRP) to explain the decision-making process of DL predictions better.
Adopting AI techniques may be beneficial for routine screening for the early identification of prevalent skin diseases. Figure 2 displays the framework of the proposed model. The proposed model has five layers and two phases: training and validation.
Framework of the proposed model.
In the training phase, layer 1 describes obtaining raw skin disease data from the open source. In layer 2, raw data is pre-processed according to the DL model. In data pre-processing, raw images acquired from open source are then converted into processed images with the RGB dimensions of 224 × 224 × 3. 224 × 224 denotes the length and width, while 3 denotes the channel count. After pre-processing, the data is randomly separated into training and testing sets for each of the four classes. For every class, 80% of the data is used for training and 20% for testing, keeping the overall dataset in the same 80:20 proportion. The pre-trained VGG16 model is imported and modified for the DL model. Layer 3 defines the predictions made by the DL model. These predictions may be perfect for decision-making, but they must explain how the DL model reached this decision. This prediction model of DL is known as the black box. To bring fairness to the decision-making process, the DL model is entangled with explainable artificial intelligence in layer 4. XAI method attempts to address the issue of opaqueness in DL models by explaining decisions based on comparisons between a model’s predictions and pre-processed data. If these explanations show any biases or inconsistencies, the model will be retrained to perform better in fairness and accuracy. When the explanations are good enough, this model will be saved on the cloud for future use. It will ensure a reliable and defensible model as this iterative process is performed.
During the validation step, which is the fifth layer of the model, the trained model is imported from the cloud to verify the pre-processed data obtained from different sources. The proposed model predicts and identifies the skin image data into four distinct classes, each with an explanation. After successfully identifying and predicting skin diseases, data is imported for future use, as shown in Fig. 2.
The study used an open-source dataset of skin diseases57. Chickenpox (107), measles (91), monkeypox (279), and normal (293) are the four classes, and a total of 770 images are present. Table 2 describes classes and the number of image samples after the data augmentation.
TL is a DL approach that uses pre-trained networks for various applications that can be used in the same domain or over different domains. The idea behind TL is to use models already trained on large and representative datasets rather than building a new CNN from scratch for each new task. With such pre-training, the first few layers of the network are being trained to extract low-level features like edges and colors, which generalize across multiple problems. It makes the learned model available for use in other applications. Based on the problem, one can fine-tune later layers of a network to adapt according to specific needs with a few more iterations trained, not the whole training again. VGG16 is utilized in this study to identify and predict skin diseases. VGG16, a deep CNN architecture with 16 layers, was constructed by Simonyan and Zisserman of the University of Oxford58. It has 16 layers that have learnable weight parameters.
The 16 layers in this proposed VGG16 architecture include 3 fully connected layers, a max-pooling layer, and 13 convolutional layers, as displayed in Fig. 3. The input layer’s images are 224 × 224 × 3 in size, and the classification layer is the last.
VGG16 original architecture59.
The VGG16 model is employed in the current study to classify four classes of skin images. Figure 3 shows the VGG16’s original architecture before modification. There are 1000 classes of different objects that the original VGG16 model was employed to classify. It is not feasible to directly use the original VGG16 mode to classify the four classes of skin images. As a result, the modification must be carried out in accordance with this article. In Fig. 4, the modified VGG16 model is shown.
Modified VGG16 architecture.
Explainability — the extent to which an AI system can act transparently and comprehensibly, not only apparently but ideally also for all60. In other words, this is just about making the process of decision-making understandable and accessible to those end-users who have no technical background. Making DL algorithms’ “black box” decision-making more transparent and intelligible is the goal of explainable artificial intelligence.
The explainability technique used in this article to explain a DL model is LRP. One major technique for explaining networks relying on the back-propagation algorithm is LRP61. This study uses the LRP technique for interpretable analysis through whether decisions made by a model reflect meaningful patterns in its input and improve generalization of proposed model. At its core, the LRP algorithm is rooted in exact reversal of contributions to tracking back from final output node layer by layer to single input nodes62. Additionally, LRP compensates for the perturbation technique (occlusion map) and the shortcoming of shattered gradients in gradient methods (Grad-CAM)63.
This article uses Google Colab and Pytorch for simulation and results. Experiment results are measured using several famous statistical metrics from Eqs. (1–8) to evaluate the proposed framework’s classification performance64. Skin conditions that are correctly diagnosed are termed true positives (Tp) or true negatives (Tn), while skin conditions that are inaccurately diagnosed are termed false positives (Fp) and false negatives (Fn). Detailed explanations of the designated statistical metrics are provided below.
Accuracy is the ratio of correctly predicted instances to the total number of instances in the dataset. It shows how the model will predict the outcome for every possible combination; thus, it is a single measure to evaluate the rate at which the model discriminates one class from another.
The Misclassification rate is the proportion of instances wrongly classified to the total number of instances. It measures the rate at which a model gives wrong predictions, shedding light on wrongly estimated outcomes.
Precision is the number of true positives divided by the sum of all false positive and true positive results. It is used to evaluate a model’s ability to accurately predict either label (positive or negative).
Specificity measures the number of instances that are actually negative out of all those predicted as negatives by the model. It would mean that the model classifies all negative instances as belonging to the negative class to prevent false positives.
Sensitivity, also known as a recall or true positive rate (TPR), measures how many of the actual positives are captured by the model. The recall score is essential because it shows how well the model can detect positive cases.
FNR measures the number of true positive instances that are classified as false negatives by the model. It shows how frequently the model misclassifies negatives; a high measure here reflects more misses on true positives.
FPR measures the ratio of actual negative cases predicted positively. It is the proportion of instances that are negative but falsely predicted as positive.
The F1 score, on the other hand, is a metric that considers both precision and sensitivity to have an overall balanced model evaluation in terms of performance under various scenarios where there are many more records per class depending on whether false positives are as important or unimportant as false negatives. The F1 score is calculated using the harmonic mean of precision and sensitivity values, which biases it to lower values.
These metrics are calculated using a confusion matrix (CM). A CM evaluates the performance of a classification model by breaking down how accurately the model makes predictions about each class. For the proposed model using the modified VGG16, the simulation was set up with a mini-batch size of 32, an optimal epoch count of 10, a learning rate of 0.00001, and the Adam optimization algorithm. The mini-batch size means the model processes 32 samples at a time to calculate gradients and update its parameters. Training the model over different epoch counts showed that 10 epochs gave the best results, with an epoch being a complete pass through the entire training dataset. The Adam optimizer, known for its efficiency and ability to handle noisy data, was used with a learning rate of 0.00001 to ensure smooth and stable training.
Testing confusion matrix for the proposed model.
Figure 5 displays the testing CM for the proposed model. A total of 462 images were utilized in the 10th epoch. In class chickenpox, 44 out of 64 images were correctly classified, while 20 images were misclassified (4 as measles, 12 as monkeypox, and 2 as normal). In the case of measles, 49 images out of 55 were correctly classified, while 6 images were misclassified (3 images as chickenpox and 3 images as monkeypox). In the case of monkeypox, 163 images out of 167 were correctly classified as monkeypox, while 4 images were misclassified as normal. In the case of class normal, 175 images out of 176 were correctly classified, while only 1 image was misclassified as monkeypox.
Table 3 provides the statistical significance of different performance metrics. It is an extensive assessment of the model’s performance with some evaluation criteria, like correctly predicting instances, minimizing error rate, and balancing trade-offs concerning false negatives vs. true positives.
Figure 6 shows how the LRP approach is applied to show why the VGG16 model gave each particular prediction. LRP is a technique that improves the interpretability of neural networks by backtracking predictions to input features, which thus represents regions in images useful for predicting with each LRP layer. Figure 6 shows the most “important” areas used by VGG16 to distinguish between classes and which regions in classes are marked using LRP. This visual representation confirms the model’s predictive performance and the reasons behind these predictions. For example, suppose the model is picking out specific components of a rash that appear to distinguish it from chickenpox. In that case, it shows how important those pieces are in allowing this image to be categorized correctly.
Explanations based on LRP for the VGG16 model.
Several methods have been utilized to identify and classify skin diseases. TL is an innovation for identifying and predicting skin diseases. It uses pre-trained models to improve specific skills. Table 4 compares the proposed model’s performance with other previously reported models. As depicted in Table 4, the proposed model competes and excels over previously reported models with a remarkable accuracy rate of 93.29%. This high accuracy underscores the proposed model’s competitive edge, convincing of its superiority in diagnosing various skin diseases. Furthermore, the proposed model incorporates the XAI technique to increase transparency and fairness.
The proposed integration of VGG16 with LRP offers practical benefits, especially in healthcare. With 93.29% accuracy, it provides a reliable tool for diagnosing skin diseases early and accurately, enhancing patient care and lowering the strain on healthcare systems. By making its predictions explainable, the model builds trust among doctors and patients, addressing the common concern of AI being a ‘black box.’ Automating disease classification can streamline workflows, lower costs, and allow clinics and telemedicine platforms to handle more patients efficiently. It also has the potential to bring accessible diagnostic tools to underserved areas through telehealth applications.
Additionally, the explainability feature aligns with ethical AI practices, helping managers and policymakers ensure transparency, compliance, and confidence in AI-based solutions. This model demonstrates how AI can transform healthcare by combining accuracy, efficiency, and trustworthiness.
The proposed model, which integrates VGG16 with LRP, achieved a notable accuracy of 93.29% and a misclassification rate of just 6.71%. These results underline the model’s ability to address the limitations of existing methods while incorporating explainability through XAI techniques. LRP enhances transparency by offering meaningful insights into the decision-making process, making it a valuable tool for building trust in AI-based systems.
At the same time, certain limitations were identified. The dataset, while suitable for the scope of this research, may only partially reflect the complexity and diversity of real-world scenarios, potentially limiting the generalizability of the findings. Additionally, privacy concerns related to sensitive data, especially in domains like healthcare, pose significant challenges. Balancing robust privacy-preserving measures with high model performance remains an ongoing priority.
Future efforts could address these limitations by testing the model on larger and more diverse datasets to ensure broader applicability across various domains. Incorporating advanced privacy-preserving approaches like federated learning or blockchain technologies could help mitigate confidentiality concerns. Further exploration of explainability techniques may also enhance the model’s interpretability, making it more accessible and transparent for both technical and non-technical users.
The dataset & Simulation files used during the current study are available from the corresponding author upon reasonable request.
Mohammed, S. S. & Al-Tuwaijari, J. M. Skin disease classification system based on machine learning technique: a Survey. IOP Conf. Ser. Mater. Sci. Eng. 1076 (012045), 1–13 (2021).
MATH Google Scholar
Al-Tbali, J., Anam, L., Al-Jamrah, K. M. & Abdul Moaen, F. Chickenpox Outbreak Investigation in Assabain District, Sana’a City, Yemen, January to February 2019, Iproceedings, vol. 8, no. 8, pp. 1–2, doi: (2022). https://doi.org/10.2196/36598
Sanjita, S., Azeem, M. & Islamovna, U. G. Survey and outbreak of chicken pox; acknowledgement by med-student, in Proceedings of the 2nd International Scientific and Practical Conference, Brussels, Belgium, pp. 77–82. (2023).
Nasiba, P. & Dildora, B. CHICKENPOX, in Proceedings of International Conference on Scientific Research in Natural and Social Sciences, Toronto, Canada, pp. 202–205. (2023).
Kujur, A., Kiran, K. A. & Kujur, M. An Epidemiological Study of Outbreak Investigation of Chickenpox in remote hamlets of a tribal state in India. Cureus 14 (6), 1–11. https://doi.org/10.7759/cureus.26454 (2022).
Article Google Scholar
Verma, R., Bairwa, M., Chawla, S., Prinja, S. & Rajput, M. Should Chickenpox vaccine be included in the national immunization schedule in India? Hum. Vaccin. 7 (8), 874–877. https://doi.org/10.4161/hv.7.8.15685 (2011).
Article PubMed Google Scholar
Chovatiya, R. & Silverberg, J. I. Inpatient morbidity and mortality of measles in the United States. PLOS ONE. 15, 1–13. https://doi.org/10.1371/journal.pone.0231329 (2020). no. 4.
Article CAS MATH Google Scholar
Rabaan, A. A. et al. Updates on measles incidence and eradication: emphasis on the immunological aspects of Measles infection. Medicina 58, 1–20. https://doi.org/10.3390/medicina58050680 (2022). no. 5.
Article Google Scholar
Gay, N. J. The theory of Measles Elimination: implications for the design of elimination strategies. J. Infect. Dis. 189, 27–35. https://doi.org/10.1086/381592 (2004).
Article MATH Google Scholar
Thornhill, J. P. et al. Monkeypox Virus infection in humans across 16 countries – April-June 2022. N Engl. J. Med. 387 (8), 679–691. https://doi.org/10.1056/NEJMoa2207323 (2022).
Article CAS PubMed MATH Google Scholar
Mitjà, O. et al. Monkeypox, Lancet, vol. 401, no. 10370, pp. 60–74, doi: (2023). https://doi.org/10.1016/S0140-6736(22)02075-X
Shchelkunov, S. N. et al. Analysis of the monkeypox virus genome. Virology 297 (2), 172–194. https://doi.org/10.1006/viro.2002.1446 (2002).
Article CAS PubMed MATH Google Scholar
Nguyen, P. Y., Ajisegiri, W., Costantino, V., Chughtai, A. A. & MacIntyre, C. R. Reemergence of human monkeypox and declining Population Immunity in the context of urbanization, Nigeria, 2017–2020. Emerg. Infect. Dis. 27 (4), 1007–1014 (2021).
Article PubMed PubMed Central Google Scholar
Doucleff, M. The spread of monkeypox was predicted by scientists in 1988: Goats and Soda : NPR. Accessed: Aug. 28, 2022. [Online]. Available: https://www.npr.org/sections/goatsandsoda/2022/05/27/1101751627/scientists-warned-us-about-monkeypox-in-1988-heres-why-they-were-right
Multi-country monkeypox outbreak in non-endemic countries. Accessed: Aug. 28. [Online]. Available: (2022). https://www.who.int/emergencies/disease-outbreak-news/item/2022-DON385
Bunge, E. M. et al. The changing epidemiology of human monkeypox—A potential threat? A systematic review. PLoS Negl. Trop. Dis. 16, 1–20. https://doi.org/10.1371/journal.pntd.0010141 (2022). no. 2.
Article MATH Google Scholar
Mansour, R. F., Althubiti, S. A. & Alenezi, F. Computer Vision with Machine Learning enabled skin lesion classification model. Comput. Mater. Contin. 73 (1), 849–864. https://doi.org/10.32604/cmc.2022.029265 (2022).
Article Google Scholar
Dong, S., Wang, P. & Abbas, K. A survey on deep learning and its applications. Comput. Sci. Rev. 40, 1–22. https://doi.org/10.1016/j.cosrev.2021.100379 (2021).
Article MathSciNet MATH Google Scholar
Abdullah, A. A., Hassan, M. M. & Mustafa, Y. T. A review on bayesian deep learning in Healthcare: Applications and challenges. IEEE Access. 10, 36538–36562. https://doi.org/10.1109/ACCESS.2022.3163384 (2022).
Article MATH Google Scholar
Yan, K., Wang, X., Lu, L. & Summers, R. M. DeepLesion: automated mining of large-scale lesion annotations and universal lesion detection with deep learning. J. Med. Imaging. 5 (03), 1–11. https://doi.org/10.1117/1.jmi.5.3.036501 (2018).
Article MATH Google Scholar
Kijowski, R., Liu, F., Caliva, F. & Pedoia, V. Deep learning for Lesion Detection, Progression, and prediction of Musculoskeletal Disease. J. Magn. Reson. Imaging. 52 (6), 1607–1619. https://doi.org/10.1002/jmri.27001 (2020).
Article PubMed MATH Google Scholar
Anupama, C. S. S. et al. Deep learning with backtracking search optimization based skin lesion diagnosis model. Comput. Mater. Contin. 70 (1), 1297–1313. https://doi.org/10.32604/cmc.2022.018396 (2021).
Article MATH Google Scholar
Talo, M., Baloglu, U. B., Yıldırım, Ö. & Rajendra Acharya, U. Application of deep transfer learning for automated brain abnormality classification using MR images. Cogn. Syst. Res. 54, 176–188. https://doi.org/10.1016/j.cogsys.2018.12.007 (2019).
Article MATH Google Scholar
Ozturk, T. et al. Automated detection of COVID-19 cases using deep neural networks with X-ray images. Comput. Biol. Med. 121, 1–11. https://doi.org/10.1016/j.compbiomed.2020.103792 (2020).
Article CAS MATH Google Scholar
Kott, O. et al. Development of a deep learning algorithm for the histopathologic diagnosis and Gleason grading of prostate Cancer biopsies: a pilot study. Eur. Urol. Focus. 7 (2), 347–351. https://doi.org/10.1016/j.euf.2019.11.003 (2021).
Article PubMed MATH Google Scholar
Shkolyar, E. et al. Augmented bladder tumor detection using deep learning. Eur. Urol. 76 (6), 714–718. https://doi.org/10.1016/j.eururo.2019.08.032 (2019).
Article PubMed PubMed Central MATH Google Scholar
Ahmed, F., Fatima, A., Mamoon, M. & Khan, S. Identification of the Diabetic Retinopathy Using ResNet-18, in 2nd International Conference on Cyber Resilience, ICCR Dubai, United Arab Emirates: IEEE, 2024, pp. 1–6. doi: (2024). https://doi.org/10.1109/ICCR61006.2024.10532925
Menaouer, B., Zoulikha, D., El-Houda, K. N., Mohammed, S. & Matta, N. Coronavirus pneumonia classification using X-Ray and CT scan images with deep convolutional neural network models. J. Inf. Technol. Res. 15 (1), 1–23. https://doi.org/10.4018/jitr.299391 (2022).
Article MATH Google Scholar
Menaouer, B., El-Houda, K. N., Zoulikha, D., Mohammed, S. & Matta, N. Detection and classification of brain tumors from MRI images using a deep convolutional neural Network Approach. Int. J. Softw. Innov. 10 (1), 1–25. https://doi.org/10.4018/IJSI.293269 (2022).
Article MATH Google Scholar
Hesamian, M. H., Jia, W., He, X. & Kennedy, P. Deep learning techniques for Medical Image Segmentation: achievements and challenges. J. Digit. Imaging. 32 (4), 582–596. https://doi.org/10.1007/s10278-019-00227-x (2019).
Article PubMed PubMed Central MATH Google Scholar
Roth, H. R. et al. Deep learning and its application to medical image segmentation. Med. IMAGING Technol. 36 (2), 63–71. https://doi.org/10.11409/mit.36.63 (2018).
Article MATH Google Scholar
Mohammed, S. S., Menaouer, B., Zohra, A. F. F. & Nada, M. Sentiment analysis of COVID-19 tweets using adaptive neuro-fuzzy inference system models. Int. J. Softw. Sci. Comput. Intell. 14 (1), 1–20. https://doi.org/10.4018/IJSSCI.300361 (2022).
Article Google Scholar
Shen, D., Wu, G. & Suk, H. I. Deep learning in Medical Image Analysis. Annu. Rev. Biomed. Eng. 176 (1), 1–35. https://doi.org/10.1146/annurev-bioeng-071516-044442.Deep (2017).
Article MATH Google Scholar
Meijering, E. A bird ’ s-eye view of deep learning in bioimage analysis. Comput. Struct. Biotechnol. J. 18, 2312–2325. https://doi.org/10.1016/j.csbj.2020.08.003 (2020).
Article CAS PubMed PubMed Central MATH Google Scholar
Jia, X., Ren, L. & Cai, J. Clinical implementation of AI technologies will require interpretable AI models. Med. Phys. 47 (1), 1–4. https://doi.org/10.1002/mp.13891 (2020).
Article CAS PubMed MATH Google Scholar
Karimkhani, C. et al. Global skin disease morbidity and mortality an update from the global burden of disease study 2013. JAMA Dermatology. 153 (5), 406–412. https://doi.org/10.1001/jamadermatol.2016.5538 (2017).
Article PubMed PubMed Central MATH Google Scholar
Seth, D., Cheldize, K., Brown, D. & Freeman, E. E. Global burden of skin disease: inequities and innovations. Curr. Dermatol. Rep. 6 (3), 204–210. https://doi.org/10.1007/s13671-017-0192-7 (2017).
Article PubMed PubMed Central MATH Google Scholar
Chang, X. & Chen, M. Research progress of varicella and its immunoprophylaxis. Front. Med. Sci. Res. 4 (5), 36–39. https://doi.org/10.25236/FMSR.2022.040507 (2022).
Article ADS MATH Google Scholar
Wutzler, P. et al. Varicella vaccination – the global experience. Expert Rev. Vaccines. 16 (8), 833–843 (2017).
Article CAS PubMed PubMed Central MATH Google Scholar
Roy, K. et al. Skin disease detection based on different segmentation techniques, in International Conference on Opto-Electronics and Applied Optics, Optronix 2019, Kolkata, India: IEEE, pp. 1–5. doi: (2019). https://doi.org/10.1109/OPTRONIX.2019.8862403
Daud, M. R. H. M., Yaacob, N. A., Ibrahim, M. I. & Muhammad, W. A. R. W. Five-Year Trend of measles and its Associated factors inPahang, Malaysia: a Population-based study. Int. J. Environ. Res. Public. Health. 19, 1–10 (2022).
MATH Google Scholar
VON MAGNUS, S., ANDERSEN, E. K., PETERSEN, K. B. & AKSEI, B. A. A POX-LIKE DISEASE IN CYNOMOLGUS MONKEYS, FROM STATENS SEHUMINSTITUT, DIRECTOH J. OHSKOV, M.D., pp. 156–176, (1959).
Ladnyj, I. D., Ziegler, P. & Kima, E. A human infection caused by monkeypox virus in Basankusu Territory, Democratic Republic of the Congo. Bull. World Health Organ. 46 (5), 593–597 (1972).
CAS PubMed PubMed Central MATH Google Scholar
Reynolds, M. G., Doty, J. B., McCollum, A. M., Olson, V. A. & Nakazawa, Y. Monkeypox re-emergence in Africa: a call to expand the concept and practice of one health. Expert Rev. Anti Infect. Ther. 17 (2), 129–139. https://doi.org/10.1080/14787210.2019.1567330 (2019).
Article CAS PubMed PubMed Central Google Scholar
Koenig, K. L., Beÿ, C. K. & Marty, A. M. Monkeypox 2022 identify-Isolate-Inform: a 3I Tool for frontline clinicians for a zoonosis with escalating human community transmission. One Heal. 15, 1–13. https://doi.org/10.1016/j.onehlt.2022.100410 (2022).
Article Google Scholar
W. H. O. (WHO), Multi-country monkeypox outbreak in non-endemic countries: Update. Accessed: Sep. 04, 2022. [Online]. Available: https://www.who.int/emergencies/disease-outbreak-news/item/2022-DON388
Ali, S. N. et al. Monkeypox Skin Lesion Detection Using Deep Learning Models: A Feasibility Study, Comput. Vis. Pattern Recognit., pp. 2–5, [Online]. Available: (2022). http://arxiv.org/abs/2207.03342
Gülmez, B. MonkeypoxHybridNet: A hybrid deep convolutional neural network model for monkeypox disease detection, Int. Res. Eng. Sci., vol. 3, pp. 49–64, [Online]. Available: (2022). https://desytamara.blogspot.com/2017/11/sistem-pelayanan-perpustakaan-dan-jenis.html%0Ahttps://lambeturah.id/pengertian-website-secara-umum-dan-menurut-para-ahli/%0Ahttps://www.researchgate.net/publication/269107473_What_is_governance/link/548173090cf2252
Irmak, M. C., Aydın, T. & Yağanoğlu, M. Monkeypox Skin Lesion Detection with MobileNetV2 and VGGNet Models, in TIPTEKNO 2022 – Medical Technologies Congress, Proceedings, Antalya, Turkey, pp. 2–5. doi: (2022). https://doi.org/10.1109/TIPTEKNO56568.2022.9960194
Singh, U. & Songare, L. S. Analysis and Detection of Monkeypox using the GoogLeNet Model, in In Proceedings of the International Conference on Automation, Computing and Renewable Systems (ICACRS), Pudukkottai, India, 2022, pp. 1000–1008. doi: (2022). https://doi.org/10.1109/ICACRS55517.2022.10029125
Sharma, K., Kishlay, V., Kumar & Mittal, M. MonkeyPox, Measles and ChickenPox Detection through Image-Processing using Residual Neural Network (ResNet), in 6th International Conference on Information Systems and Computer Networks, ISCON 2023, Mathura, India: IEEE, 2023, pp. 1–6. doi: (2023). https://doi.org/10.1109/ISCON57294.2023.10112085
Sethy, P. K. et al. Detection of Monkeypox Based on Improved Darknet19, in IEEE 8th International Conference for Convergence in Technology, I2CT 2023, Pune, India: IEEE, 2023, pp. 1–3. doi: (2023). https://doi.org/10.1109/I2CT57861.2023.10126170
Uysal, F. Detection of Monkeypox Disease from Human skin images with a Hybrid Deep Learning Model. Diagnostics 13 (10), 1–23. https://doi.org/10.3390/diagnostics13101772 (2023).
Article MATH Google Scholar
Ariansyah, M. H., Winarno, S. & Sani, R. R. Monkeypox and Measles Detection using CNN with VGG-16 transfer learning. J. Comput. Res. Innov. 8 (1), 32–44. https://doi.org/10.3390/s23041783 (2023).
Article Google Scholar
Kundu, D., Siddiqi, U. R. & Rahman, M. M. Vision Transformer based Deep Learning Model for Monkeypox Detection, in 25th International Conference on Computer and Information Technology (ICCIT), Cox’s Bazar, Bangladesh: IEEE, pp. 1021–1026. doi: (2023). https://doi.org/10.1109/iccit57492.2022.10054797
Akram, A. et al. SkinMarkNet: an automated approach for prediction of monkeyPox using image data augmentation with deep ensemble learning models. Multimed Tools Appl. 1–17. https://doi.org/10.1007/s11042-024-19862-w (2024).
Monkeypox Skin Images Dataset (MSID). | Kaggle. Accessed: Aug. 28, 2022. [Online]. Available: https://www.kaggle.com/datasets/dipuiucse/monkeypoxskinimagedataset
Simonyan, K. & Zisserman, A. Very deep convolutional networks for large-scale image recognition, in Published as a conference paper at ICLR, pp. 1–14. (2015).
Althubiti, S. A., Alenezi, F., Shitharth, S., Sangeetha, K. & Reddy, C. V. S. Circuit Manufacturing Defect Detection Using VGG16 Convolutional Neural Networks, Wirel. Commun. Mob. Comput., vol. pp. 1–10, 2022, doi: (2022). https://doi.org/10.1155/2022/1070405
Doshi-Velez, F. & Kim, B. Towards a Rigorous Science of interpretable machine learning. arXiv Prepr, pp. 1–13, (2017).
Bach, S. et al. On pixel-wise explanations for non-linear classifier decisions by layer-wise relevance propagation. PLoS One. 10 (7), 1–46. https://doi.org/10.1371/journal.pone.0130140 (2015).
Article CAS MATH Google Scholar
Böhle, M., Eitel, F., Weygandt, M. & Ritter, K. Layer-wise relevance propagation for explaining deep neural network decisions in MRI-based Alzheimer’s disease classification. Front. Aging Neurosci. 11, 1–17. https://doi.org/10.3389/fnagi.2019.00194 (2019).
Article MATH Google Scholar
Huang, X., Jamonnak, S., Zhao, Y., Wu, T. H. & Xu, W. A visual designer of layer-wise relevance propagation models. Eurographics Conf. Vis. 40 (3), 227–238 (2021).
Google Scholar
Seliya, N., Khoshgoftaar, T. M. & Van Hulse, J. A study on the relationships of classifier performance metrics, in 21st IEEE International Conference on Tools with Artificial Intelligence, Newark, NJ, USA, pp. 59–66. doi: (2009). https://doi.org/10.1109/ICTAI.2009.25
Download references
This research work is supported by Prince Mohammad Bin Fahd University, Al-Khobar, Dhahran, 34754, Saudi Arabia.
Department of Computer Science, Prince Mohammad Bin Fahd University, 34754, Al-Khobar, Dhahran, KSA, Saudi Arabia
Sagheer Abbas
School of Computer Science, National College of Business Administration and Economics, Lahore, 54000, Pakistan
Fahad Ahmed & Munir Ahmad
Department of Computer Science, Baba Guru Nanak University, Nankana Sahib, 39100, Pakistan
Wasim Ahmad Khan
College of Informatics, Korea University, Seoul, 02841, Republic of Korea
Munir Ahmad
Department of Software, Faculty of Artificial Intelligence and Software, Gachon University, Seongnam-si, 13120, Republic of Korea
Muhammad Adnan Khan
Research Innovation and Entrepreneurship Unit, University of Buraimi, 512, Buraimi, Oman
Taher M. Ghazal
Center for Cyber Security, Faculty of Information Science and Technology, Universiti Kebangsaan Malaysia (UKM), Bangi, 43600, Selangor, Malaysia
Taher M. Ghazal
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
S.A., F.A., W.A.K., and M.A; have collected data from different resources, contributed in writing—original draft preparation, drafted pictures and tables. S.A., F.A., M.A., and M.A.K; performed formal analysis and Simulation, performed revision and improve the quality of the draft. F.A., T.M.G., W.A.K, and M.A.K.; writing—review and editing, performed supervision. All authors have read and agreed to the published version of the manuscript.
Correspondence to Muhammad Adnan Khan.
The authors declare no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and permissions
Abbas, S., Ahmed, F., Khan, W.A. et al. Intelligent skin disease prediction system using transfer learning and explainable artificial intelligence. Sci Rep 15, 1746 (2025). https://doi.org/10.1038/s41598-024-83966-4
Download citation
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598-024-83966-4
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
Advertisement
© 2025 Springer Nature Limited
Sign up for the Nature Briefing: Cancer newsletter — what matters in cancer research, free to your inbox weekly.
