Automated Detection and Classification of Pleural Effusion on Computed Tomography Using Deep Learning

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Pleural effusion diagnosis requires accurate segmentation and etiological classification.
• Current methods can be time-consuming and subjective.
• Noncontrast thoracic computed tomography (CT) is widely used for imaging.

Purpose of the Study:

• To develop and evaluate a two-stage deep learning AI framework.
• To achieve automatic segmentation of pleural effusion.
• To classify pleural effusion etiology (empyema, malignant, transudative).

Main Methods:

• Retrospective study using noncontrast thoracic CT images.
• Stage 1: U-Net deep learning for pleural effusion segmentation.
• Stage 2: Classification using quantitative features (area, density, texture) and machine learning models (logistic regression, SVM, random forest, gradient boosting).

Main Results:

• The U-Net model achieved high segmentation performance.
• Gradient boosting and random forest models yielded 96% accuracy and 0.95 macro F1-score for three-class etiological discrimination.
• Key discriminative features included effusion area, intensity standard deviation, and texture heterogeneity (GLCM).

Conclusions:

• The two-stage AI framework accurately segments and classifies pleural effusion from noncontrast thoracic CT.
• The system demonstrates potential as a clinical decision support tool.
• It enables rapid, objective, and reproducible evaluation of pleural effusions.