Driverless artificial intelligence framework for the identification of malignant pleural effusion.

Area of Science:

• Medical Artificial Intelligence
• Machine Learning in Medicine
• Diagnostic Accuracy

Background:

• Distinguishing malignant pleural effusion (MPE) from benign pleural effusion (BPE) is clinically challenging.
• Accurate differentiation is crucial for appropriate patient management and treatment strategies.

Purpose of the Study:

• To evaluate the efficacy of deep learning (DL) and machine learning (ML) models in differentiating MPE from BPE.
• To assess the external validity of these predictive models.

Main Methods:

• Utilized a retrospective cohort of 726 patients with pleural effusion (PE) for model training and testing.
• Employed driverless AI, DL, and five ML models: GBM, XGBoost, XRT, DRF, and GLM.
• Validated the models using a prospective cohort of 172 PE patients.

Main Results:

• The stacked ensemble AI model achieved high diagnostic performance with Area Under the Curve (AUC) of 0.991 (training), 0.912 (test), and 0.953 (validation).
• Gradient Boosting Machine (GBM) demonstrated robust and consistent predictive efficiency across all datasets.
• While DL excelled in the training set (AUC=0.995), GBM showed superior stability in external validation.

Conclusions:

• Driverless AI, leveraging routinely collected clinical data, significantly enhances diagnostic performance for MPE vs. BPE.
• Machine learning models, especially GBM, offer a promising and reliable approach for pleural effusion classification.
• The developed AI framework holds potential for improving clinical decision-making in diagnosing pleural effusions.