SOSPCNN: Structurally Optimized Stochastic Pooling Convolutional Neural Network for Tetralogy of Fallot recognition

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Cardiology

Background:

• Tetralogy of Fallot (TOF) is a complex congenital heart disease requiring accurate and timely diagnosis.
• Current diagnostic methods for TOF can be time-consuming and may lack precision.
• Advancements in artificial intelligence offer potential for improving diagnostic accuracy in cardiovascular diseases.

Purpose of the Study:

• To develop and evaluate a novel artificial intelligence model for efficient and precise recognition of Tetralogy of Fallot (TOF).
• To leverage cardiovascular computed tomography (CT) imaging for AI-driven TOF diagnosis.

Main Methods:

• Development of a structurally optimized stochastic pooling convolutional neural network (SOSPCNN).
• Integration of stochastic pooling, structural optimization, and convolutional neural network architectures.
• Utilization of multiple-way data augmentation to prevent overfitting and Grad-CAM for model explainability.

Main Results:

• The SOSPCNN model achieved high performance metrics, including 92.50% accuracy and 0.9587 AUC.
• Sensitivity, specificity, and precision were all above 92%, demonstrating robust diagnostic capability.
• The model outperformed three existing state-of-the-art TOF recognition methods in evaluations.

Conclusions:

• The proposed SOSPCNN model demonstrates significant potential for accurate and efficient Tetralogy of Fallot diagnosis.
• The AI-driven approach offers a valuable tool for clinicians in cardiovascular imaging.
• Development of associated desktop and web applications facilitates clinical implementation.