AI-Powered Monkeypox Diagnosis: Optimizing Accuracy and Speed

By Dr Silpaja Chandrasekar, PhDReviewed by Susha Cheriyedath, M.Sc.Sep 6 2023

In a paper published in the journal Scientific Reports, researchers introduced a Convolutional Neural Networks (CNN)-based method for classifying monkeypox skin lesions. They optimized the model using the Grey Wolf Optimizer (GWO), significantly improving accuracy. This method can make monkeypox diagnosis faster and more accurate. It helps early disease detection, improves patient outcomes, and also has the potential to benefit public health by controlling outbreaks.

Background

Monkeypox is caused by the monkeypox virus and is primarily found in Central and West Africa, with occasional outbreaks worldwide. Diagnosis relies on clinical symptoms and lab tests, which include Polymerase Chain Reaction (PCR) for virus detection. Artificial intelligence (AI), particularly CNN, has shown promise in skin disease diagnosis but also faces challenges due to limited data, variable lesion characteristics, potential misclassification, resource requirements, and validation needs.

Related work

Prior research has explored the global impact of a multinational monkeypox outbreak with over 4900 cases reported by June 2022 across the Western Hemisphere. The diagnosis of monkeypox depends on clinical evaluation, examination of skin lesions, and PCR testing, with the absence of a specific treatment at present. The use of CNN has demonstrated promise in the diagnosis of skin diseases. Researchers have proposed various AI models for monkeypox detection, such as ANN-based systems and deep CNNs like "MonkeyNet", which has achieved high accuracy. Metaheuristic algorithms like the Al-Biruni Earth radius (BER) optimization technique have also been explored for classification tasks. 

Proposed method

The Monkeypox prediction model comprises four phases: data preprocessing, feature selection, Monkeypox prediction with a CNN model, and CNN hyperparameter optimization using the GWO algorithm. The Monkeypox dataset was collected during the 2022 outbreak in London and includes 25,000 instances and 11 features with the presence or absence of monkeypox. After preprocessing and balancing, feature selection identifies significant symptoms. A CNN architecture is used for Monkeypox prediction, with various CNN architectures evaluated. Finally, the GWO algorithm optimizes CNN hyperparameters for enhanced accuracy.

Experimental analysis

Experiments were conducted to evaluate the performance of a monkeypox prediction model using a 3 GHz AMD Ryzen 7 computer with 16 GB RAM and Python. The focus was on data preprocessing, feature selection, and hyperparameter optimization.

Experiment I: The CNN model was initially run without preprocessing or optimization, resulting in moderate performance metrics.

Experiment II: The dataset was improved by handling missing values, balancing data with SMOTEEN, and selecting key features. The CNN hyperparameters were optimized using the GWO algorithm, and this approach significantly enhanced model performance.

The GWO algorithm was applied with various settings, leading to varying hyperparameters and fitness scores. Notably, a population size of 50 and 10 iterations yielded the best results and achieved high accuracy (95.312%) and other favorable metrics.

The CNN model with GWO optimization outperformed the one without in terms of accuracy, precision, recall, F1 score, and AUC score. The GWO optimization significantly improved the AUC score, increasing it from 61.475 to 92.686, which indicates a substantial enhancement in classifying positive and negative cases. Overall, GWO optimization proved valuable for enhancing CNN model performance in monkeypox prediction tasks.

Contribution of this paper

The contribution of this paper can be summarized as follows:

Development of a Monkeypox Prediction Model: The paper presents the development of a prediction model for monkeypox using a CNN approach, which addresses a crucial healthcare challenge.

Data Preprocessing: It highlights the importance of data preprocessing, including handling missing values and balancing the dataset using SMOTEEN.

Feature Selection: The paper explores feature selection techniques to identify the most significant factors affecting monkeypox diagnoses by providing insights into the key clinical features.

Hyperparameter Optimization: It introduces the use of the GWO algorithm for automating the optimization of CNN hyperparameters to reduce the need for manual tuning.

Performance Improvement: The study demonstrates that GWO optimization significantly improves the model's performance in terms of accuracy, precision, recall, F1 score, and Area Under the Curve (AUC) score. It is more effective in classifying positive and negative cases.

Clinical Relevance: This paper's findings have practical implications for healthcare professionals by potentially aiding in the early and accurate diagnosis of monkeypox, contributing to better patient outcomes.

Methodological Insights: It provides insights into the methodology of employing deep learning techniques, data preprocessing, and hyperparameter optimization for disease prediction and can be applied to other medical domains.

Conclusion

In conclusion, this paper has presented a comprehensive approach to monkeypox prediction by emphasizing the significance of data preprocessing, feature selection, and GWO optimization. The study's findings have demonstrated substantial improvements in model performance and also highlighted its potential to assist healthcare professionals in early and accurate monkeypox diagnoses. The model's methodology and insights gained from this research can be applied to other medical domains. This method has contributed to the broader field of disease prediction and healthcare advancements.

In future work, several promising avenues exist for further enhancing the monkeypox prediction model. Expanding the dataset to include a more extensive and diverse set of clinical features could lead to even more accurate predictions. Additionally, exploring advanced optimization techniques beyond GWO may yield further improvements. Real-world deployment and collaboration with healthcare experts to integrate the model into clinical practice are essential to validate its practical utility and effectiveness in diagnosing monkeypox cases. Moreover, efforts should be made to ensure the interpretability of the model and address ethical concerns related to patient data.