[Paper Review] Monkeypox Skin Lesion Detection Using Deep Learning Models: A Feasibility Study

Shams Nafisa Ali, Md. Tazuddin Ahmed+5 more|arXiv (Cornell University)|Jul 6, 2022

Poxvirus research and outbreaks126 citations

TL;DR

The paper introduces the Monkeypox Skin Lesion Dataset (MSLD) and evaluates transfer-learning CNNs (VGG16, ResNet50, InceptionV3) for monkeypox detection via skin lesions, achieving best accuracy with ResNet50 in a 3-fold cross-validation. It also presents a prototype web-tool for screening.

ABSTRACT

The recent monkeypox outbreak has become a public health concern due to its rapid spread in more than 40 countries outside Africa. Clinical diagnosis of monkeypox in an early stage is challenging due to its similarity with chickenpox and measles. In cases where the confirmatory Polymerase Chain Reaction (PCR) tests are not readily available, computer-assisted detection of monkeypox lesions could be beneficial for surveillance and rapid identification of suspected cases. Deep learning methods have been found effective in the automated detection of skin lesions, provided that sufficient training examples are available. However, as of now, such datasets are not available for the monkeypox disease. In the current study, we first develop the ``Monkeypox Skin Lesion Dataset (MSLD)" consisting skin lesion images of monkeypox, chickenpox, and measles. The images are mainly collected from websites, news portals, and publicly accessible case reports. Data augmentation is used to increase the sample size, and a 3-fold cross-validation experiment is set up. In the next step, several pre-trained deep learning models, namely, VGG-16, ResNet50, and InceptionV3 are employed to classify monkeypox and other diseases. An ensemble of the three models is also developed. ResNet50 achieves the best overall accuracy of $82.96(\pm4.57\%)$, while VGG16 and the ensemble system achieved accuracies of $81.48(\pm6.87\%)$ and $79.26(\pm1.05\%)$, respectively. A prototype web-application is also developed as an online monkeypox screening tool. While the initial results on this limited dataset are promising, a larger demographically diverse dataset is required to further enhance the generalizability of these models.

Motivation & Objective

• Create an open Monkeypox Skin Lesion Dataset (MSLD) containing monkeypox, chickenpox, and measles images.
• Assess the feasibility of transfer-learning CNNs for binary monkeypox vs. others skin lesion classification.
• Evaluate three pretrained models (VGG16, ResNet50, InceptionV3) and an ensemble for accuracy and consistency.
• Develop a prototype web-application for remote monkeypox screening based on the models.

Proposed method

• Assemble and pre-process the Monkeypox Skin Lesion Dataset (MSLD) with 228 original images and 3192 augmented images.
• Apply data augmentation (rotation, translation, reflection, hue, saturation, brightness, noise, blur, scaling).
• Fine-tune pretrained CNNs (VGG16, ResNet50, InceptionV3) by unfreezing the bottom eight layers and adding three FC layers with dropout.
• Train with Adam optimizer (lr=1e-5), binary cross-entropy loss, batch size 16, on 224x224x3 inputs with 3-fold cross-validation (70/10/20 split).
• Evaluate using accuracy, precision, F1-score, and sensitivity; report mean and standard deviation across folds.
• Deploy the best model in a prototype web-app for user-uploaded lesion analysis.

Experimental results

Research questions

• RQ1 Can a public Monkeypox Skin Lesion Dataset (MSLD) enable machine learning-based screening for monkeypox from skin lesions?
• RQ2 Do pretrained CNNs (VGG16, ResNet50, InceptionV3) with transfer learning achieve reliable binary classification (monkeypox vs others) on the MSLD?
• RQ3 Is an ensemble of the three models more robust or consistent across folds?
• RQ4 What is the feasibility of a web-based screening tool using these models for remote preliminary assessment?

Key findings

• ResNet50 achieved the highest accuracy: 82.96% (±4.57).
• VGG16 achieved 81.48% (±6.87) accuracy.
• InceptionV3 achieved 74.07% (±3.78) accuracy.
• Ensemble achieved 79.26% (±1.05) accuracy, with the lowest std dev across folds.
• The study notes limited dataset size and demographic diversity as major generalization constraints.
• A prototype web-app for online screening was developed.