Related Experiment Video
Updated: Sep 21, 2025
08:05
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
14.3K
Automated Detection, Segmentation, and Classification of Pericardial Effusions on Chest CT Using a Deep Convolutional
Adrian Jonathan Wilder-Smith1,2, Shan Yang1, Thomas Weikert1,2
1Division of Research and Analytical Services, University Hospital Basel, 4031 Basel, Switzerland.
Diagnostics (Basel, Switzerland)
|May 28, 2022
Summary
An AI tool accurately detects, segments, and classifies pericardial effusions (PEFs) on CT scans, improving diagnosis for patients with hemodynamic compromise. This automated system enhances reporting and aids in critical patient care.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Radiology
- Cardiovascular Imaging
Background:
- Pericardial effusions (PEFs) are frequently missed on Computed Tomography (CT) scans.
- Missed PEFs can negatively impact patient outcomes, especially in cases of hemodynamic compromise.
- Automated tools are needed to improve the accuracy and efficiency of PEF diagnosis on CT.
Purpose of the Study:
- To develop and validate an automated deep learning tool for detecting, segmenting, and classifying pericardial effusions on CT.
- To assess the performance of the automated tool against manual segmentation and in external datasets.
- To evaluate the tool's utility in improving CT-based PEF diagnosis and reporting.
Main Methods:
- A deep convolutional neural network (nnU-Net) was trained on 316 CT scans (with and without PEF, including simple PEF and hemopericardium).
- The model was tested on 200 internal and 22 external post-mortem CT scans.
- Performance metrics included sensitivity, specificity, Area Under the Curve (AUC), and Dice coefficient; classification used median Hounsfield unit.
- Inter-reader variability was assessed on 40 CT scans.
Main Results:
- The model achieved 97% sensitivity and 100% specificity for PEF detection.
- It demonstrated 89.74% sensitivity and 83.61% specificity for hemopericardium diagnosis (AUC 0.944).
- Model performance (Dice 0.75) was superior to inter-reader variability (Dice 0.69) and unaffected by contrast or other chest pathologies.
- External validation confirmed similar performance.
Conclusions:
- The developed AI model reliably detects, segments, and classifies pericardial effusions on CT.
- This tool has the potential to serve as an alert system, enhancing the quality of radiology reports.
- The publicly available model and datasets can advance automated PEF diagnosis in clinical practice.
