PENet: Continuous-Valued Pulmonary Edema Severity Prediction On Chest X-ray Using Siamese Convolutional Networks

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate assessment of pulmonary edema severity in chest radiographs is crucial for timely clinical decisions in congestive heart failure management.
• Current deep learning models show limitations in predicting continuous-valued severity of lung edema, despite success in detecting its presence or discrete grades.

Purpose of the Study:

• To introduce PENet, a novel Siamese convolutional neural network designed for assessing the continuous spectrum of lung edema severity from chest radiographs.
• To evaluate different implementation modes of PENet and compare its performance against existing methods.

Main Methods:

• Development of PENet, a Siamese convolutional neural network architecture.
• Implementation and testing of various network configurations.
• Performance evaluation using chest radiographs with a focus on continuous severity prediction.

Main Results:

• The best performing PENet model achieved a mean Area Under the Curve (AUC) of 0.91, outperforming previous work (0.87).
• The model demonstrated high efficiency, utilizing only 1/16th the dimension of input images and 1/69th the size of training data compared to prior methods.
• Significant reduction in computational cost was observed.

Conclusions:

• PENet offers a promising solution for continuous lung edema severity assessment in chest radiographs.
• The model's efficiency in terms of data and computational requirements makes it a practical tool for clinical application.
• PENet has the potential to enhance clinical decision-making for patients with congestive heart failure.