[Paper Review] Image Data collection and implementation of deep learning-based model in detecting Monkeypox disease using modified VGG16

Md Manjurul Ahsan, Muhammad Ramiz Uddin+4 more|arXiv (Cornell University)|Jun 4, 2022

Poxvirus research and outbreaks88 citations

TL;DR

This paper presents the Monkeypox2022 image dataset and a modified VGG16 model for Monkeypox detection from skin images, with evaluation on two study setups and explainability via LIME.

ABSTRACT

While the world is still attempting to recover from the damage caused by the broad spread of COVID-19, the Monkeypox virus poses a new threat of becoming a global pandemic. Although the Monkeypox virus itself is not deadly and contagious as COVID-19, still every day, new patients case has been reported from many nations. Therefore, it will be no surprise if the world ever faces another global pandemic due to the lack of proper precautious steps. Recently, Machine learning (ML) has demonstrated huge potential in image-based diagnoses such as cancer detection, tumor cell identification, and COVID-19 patient detection. Therefore, a similar application can be adopted to diagnose the Monkeypox-related disease as it infected the human skin, which image can be acquired and further used in diagnosing the disease. Considering this opportunity, in this work, we introduce a newly developed "Monkeypox2022" dataset that is publicly available to use and can be obtained from our shared GitHub repository. The dataset is created by collecting images from multiple open-source and online portals that do not impose any restrictions on use, even for commercial purposes, hence giving a safer path to use and disseminate such data when constructing and deploying any type of ML model. Further, we propose and evaluate a modified VGG16 model, which includes two distinct studies: Study One and Two. Our exploratory computational results indicate that our suggested model can identify Monkeypox patients with an accuracy of $97\pm1.8\%$ (AUC=97.2) and $88\pm0.8\%$ (AUC=0.867) for Study One and Two, respectively. Additionally, we explain our model's prediction and feature extraction utilizing Local Interpretable Model-Agnostic Explanations (LIME) help to a deeper insight into specific features that characterize the onset of the Monkeypox virus.

Motivation & Objective

• Create a publicly available Monkeypox image dataset sourced from open portals to enable ML research.
• Develop a low-modified VGG16 transfer-learning model to distinguish Monkeypox from non-Monkeypox images.
• Provide explainability of model predictions using LIME to identify discriminative image regions.
• Evaluate model performance on two dataset setups to assess feasibility with small to moderate data.
• Highlight limitations and provide directions for future data collection and model deployment.

Proposed method

• Assemble the Monkeypox2022 dataset from open-source images including Monkeypox, Chickenpox, Measles, and Normal skin images; augment data using Keras ImageDataGenerator with rotation, shifts, zoom, shear, flipping, and reflective filling.
• Implement a modified VGG16 architecture with added updated layers and dense/dropout blocks for binary classification (Monkeypox vs Others).
• Tune hyperparameters via grid search for two studies; Study One uses smaller batches and epochs, Study Two uses larger batches and epochs with Adam optimizer.
• Apply Local Interpretable Model-Agnostic Explanations (LIME) to generate image-level explanations and identify salient regions contributing to predictions.
• Evaluate performance using accuracy, precision, recall, F1-score, sensitivity, specificity, and AUC; report results with 95% confidence intervals due to small sample sizes.
• Datasets split 80/20 into train/test; Study One: Monkeypox 43 total (train 34, test 9) and Others 47 total (train 38, test 9); Study Two: Monkeypox augmented 587 (train 470, test 117) and Others 1167 (train 933, test 234).

Experimental results

Research questions

• RQ1 Can a publicly available Monkeypox image dataset enable effective ML-based detection of Monkeypox from skin images?
• RQ2 How does a modified VGG16 model perform in distinguishing Monkeypox from non-Monkeypox images on small-to-moderate datasets?
• RQ3 Do transfer-learning approaches with data augmentation yield reliable performance in early Monkeypox image classification research?
• RQ4 Can LIME explanations reveal anatomically plausible regions driving Monkeypox predictions to support model interpretability?
• RQ5 What are the limitations and practical considerations for deploying such models in real-time screening or mobile tools?

Key findings

• Study One achieved high training accuracy (0.97) with strong AUC (0.972) but lower test performance (accuracy 0.83, AUC 0.972 0.748 for ROC curves reported).
• Study Two yielded lower overall performance due to dataset imbalance (Monkeypox vs Others) with train accuracy 0.88 and test accuracy 0.78; AUC on test set 0.748.
• LIME visualization highlighted regions driving predictions and helped interpret misclassifications (e.g., Study One misclassified Monkeypox as Chickenpox due to focus on non-symptomatic regions).
• Transfer-learning with a modified VGG16 can achieve 78%–97% accuracy across small-to-moderate datasets, indicating feasibility of image-based Monkeypox screening with limited data.
• Authors provide a publicly available Monkeypox dataset to spur further research and comparisons as data collection scales up.
• The study emphasizes explainability (LIME) as a requirement for clinical trial readiness and potential real-time screening applications.