Confidence-Aware Severity Assessment of Lung Disease from Chest X-Rays Using Deep Neural Network on a Multi-Reader...

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Radiology

Background:

• Accurate severity classification of lung diseases in COVID-19 patients using chest X-rays (CXRs) is crucial for patient management.
• Existing methods may lack confidence measures, limiting their clinical utility.
• Bayesian neural networks (BNNs) offer a probabilistic approach to enhance diagnostic models.

Purpose of the Study:

• To develop and evaluate a confidence-aware severity classification model for COVID-19 lung disease using CXRs.
• To approximate Bayesian neural networks (BNNs) using Monte Carlo Dropout (MCD) for this task.
• To assess the model's performance against human readers and its generalization capabilities.

Main Methods:

• A Monte Carlo Dropout (MCD) based method was employed as a Bayesian neural network (BNN) approximation.
• The model was trained and tested on 1208 CXRs, classifying severity into normal, mild, moderate, and severe categories.
• Internal validation and external testing on diverse datasets (2200 and 1300 CXRs) were performed.

Main Results:

• The model achieved an average area under the curve (AUC) of 0.94 ± 0.01 on the primary dataset.
• It surpassed human readers in each severity class, with a Kendall correlation coefficient (KCC) of 0.80 ± 0.03.
• Consistent performance across varied datasets demonstrated strong generalization capabilities.

Conclusions:

• The developed BNN model effectively classifies COVID-19 lung disease severity from CXRs, outperforming human readers.
• Predictive uncertainty (PU) measures correlate with radiologist agreement, aiding in clinical decision-making.
• The model's confidence-aware predictions enhance diagnostic precision and show potential for clinical application.