Artificial intelligence-based pulmonary embolism classification: Development and validation using real-world data.

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Pulmonary embolism (PE) is a critical condition requiring timely diagnosis.
• Computed tomography angiography (CTA) is a primary imaging modality for PE detection.
• Accurate and efficient AI tools are needed to aid in PE diagnosis.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) based classification model for detecting pulmonary embolism (PE) in computed tomography angiography (CTA).
• To assess the model's performance at both slice and examination levels, and its ability to exclude PE cases.

Main Methods:

• A two-dimensional AI model integrating temporal series was developed using a two-stage training process: Convolutional Neural Network (CNN) InceptionResNet V2 followed by Recurrent Neural Network (RNN) Long Short-Term Memory (LSTM).
• The model was trained on public data and validated on a large tertiary hospital dataset, including external validation.
• Performance metrics included accuracy, precision, and recall for both PE detection and exclusion.

Main Results:

• The AI model achieved 93% accuracy at the slice level and 77% at the examination level.
• External validation showed 86% precision for positive PE cases and 69% for negative cases.
• The model demonstrated strong performance in excluding PE, with 73% precision and 82% recall, highlighting its clinical utility.

Conclusions:

• The AI-based classification model shows significant promise for accurate detection and exclusion of pulmonary embolism in CTA.
• The model's generalizability is supported by validation on a diverse demographic dataset.
• This AI tool has the potential to streamline PE diagnosis, reduce unnecessary interventions, and improve patient outcomes.