Automated Detection, Segmentation, and Classification of Pleural Effusion From Computed Tomography Scans Using...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging Analysis

Background:

• Pleural effusions are common findings on chest CT scans.
• Accurate detection, segmentation, and classification of pleural effusions are crucial for patient management.
• Distinguishing between simple and complex effusions can be challenging.

Purpose of the Study:

• To develop and evaluate artificial intelligence (AI) algorithms for automated detection, segmentation, and classification of pleural effusions on computed tomography (CT) scans.
• To assess the performance of these algorithms in differentiating simple from complex effusions.

Main Methods:

• A deep convolutional neural network (nnU-Net) was trained for effusion detection and segmentation using 160 positive and 160 negative chest CT scans.
• A random forest model was implemented for classification of segmented effusions using radiomics-based features on a separate cohort of 335 patients.
• Performance was evaluated using sensitivity, specificity, AUC, Dice coefficient, and volume difference.

Main Results:

• The AI model achieved excellent sensitivity (0.99) and specificity (0.98) for effusion detection.
• Segmentation performance was robust with a median Dice coefficient of 0.89 and median absolute volume difference of 13 mL.
• Classification of simple versus complex effusions yielded a sensitivity of 0.67, specificity of 0.75, and AUC of 0.77.

Conclusions:

• Robust AI models were developed for the detection, segmentation, and classification of pleural effusion subtypes.
• The developed algorithms demonstrate potential for improving the analysis of pleural effusions in clinical practice.
• The algorithms are openly available to facilitate further research and application.