[Paper Review] UncertaintyFuseNet: Robust Uncertainty-aware Hierarchical Feature Fusion Model with Ensemble Monte Carlo Dropout for COVID-19 Detection

Moloud Abdar, Soorena Salari+8 more|arXiv (Cornell University)|May 18, 2021

COVID-19 diagnosis using AI4 citations

TL;DR

This paper proposes UncertaintyFuseNet, a deep learning model that fuses hierarchical features from chest CT and X-ray images using ensemble Monte Carlo dropout for uncertainty-aware COVID-19 detection. It achieves 99.08% accuracy on CT scans and 96.35% on X-rays, demonstrating superior robustness to noise and out-of-distribution data.

ABSTRACT

The COVID-19 (Coronavirus disease 2019) pandemic has become a major global threat to human health and well-being. Thus, the development of computer-aided detection (CAD) systems that are capable to accurately distinguish COVID-19 from other diseases using chest computed tomography (CT) and X-ray data is of immediate priority. Such automatic systems are usually based on traditional machine learning or deep learning methods. Differently from most of existing studies, which used either CT scan or X-ray images in COVID-19-case classification, we present a simple but efficient deep learning feature fusion model, called UncertaintyFuseNet, which is able to classify accurately large datasets of both of these types of images. We argue that the uncertainty of the model's predictions should be taken into account in the learning process, even though most of existing studies have overlooked it. We quantify the prediction uncertainty in our feature fusion model using effective Ensemble MC Dropout (EMCD) technique. A comprehensive simulation study has been conducted to compare the results of our new model to the existing approaches, evaluating the performance of competing models in terms of Precision, Recall, F-Measure, Accuracy and ROC curves. The obtained results prove the efficiency of our model which provided the prediction accuracy of 99.08\% and 96.35\% for the considered CT scan and X-ray datasets, respectively. Moreover, our UncertaintyFuseNet model was generally robust to noise and performed well with previously unseen data. The source code of our implementation is freely available at: https://github.com/moloud1987/UncertaintyFuseNet-for-COVID-19-Classification.

Motivation & Objective

• Address the lack of uncertainty quantification in existing deep learning models for COVID-19 detection using medical imaging.
• Develop a unified, robust feature fusion model capable of processing both CT and X-ray images for improved diagnostic accuracy.
• Enhance model reliability by integrating uncertainty estimation into the training and inference process.
• Ensure generalization and robustness to noise and unseen data in clinical deployment scenarios.
• Provide a scalable, generalizable framework applicable to other complex disease detection tasks.

Proposed method

• Employ a hierarchical feature fusion architecture that combines low-level and high-level features from pre-trained CNNs on CT and X-ray images.
• Integrate Ensemble Monte Carlo Dropout (EMCD) to estimate model uncertainty by sampling multiple forward passes with dropout active during inference.
• Use early fusion strategy to merge features from different imaging modalities before final classification.
• Train the model using cross-entropy loss with weight regularization and data augmentation to improve generalization.
• Apply class-wise analysis and robustness evaluation to validate consistent performance across all classes.
• Utilize a two-branch encoder structure with shared and modality-specific layers to extract modality-specific representations before fusion.

Experimental results

Research questions

• RQ1 Can a unified deep learning model effectively fuse features from both CT and X-ray images to improve COVID-19 detection accuracy?
• RQ2 How does incorporating uncertainty quantification via Ensemble MC Dropout enhance model reliability and interpretability in medical diagnosis?
• RQ3 To what extent is the proposed model robust to noise and capable of detecting unknown or out-of-distribution samples?
• RQ4 How does UncertaintyFuseNet compare to state-of-the-art models in terms of precision, recall, F-measure, and AUC-ROC on standard benchmark datasets?
• RQ5 Can the model generalize across different imaging modalities and maintain high performance without fine-tuning?

Key findings

• UncertaintyFuseNet achieved 99.08% accuracy on the CT scan dataset, significantly outperforming all baseline models including ResNet50 and DeTraC.
• On the X-ray dataset, the model reached 96.35% accuracy, surpassing state-of-the-art models such as DarkCovidNet and Bayesian Deep Learning.
• The model demonstrated high robustness to noise, maintaining strong performance even under adverse data conditions.
• The integration of Ensemble MC Dropout enabled reliable uncertainty estimation, allowing the model to detect unknown or anomalous cases with high confidence.
• Class-wise analysis confirmed consistent performance across all diagnostic categories, indicating balanced and reliable predictions.
• The model’s performance was validated using comprehensive metrics including precision, recall, F-measure, and ROC-AUC, all of which were superior to existing methods.