Innovative Use of Convolutional Neural Networks in Coronary Angiography
Cardiac index (CI) is a critical measurement in evaluating the heart’s performance, particularly in patients undergoing coronary angiography (CAG). Traditionally, obtaining CI data requires an invasive procedure, which can deter its frequent use. However, a recent study explores a groundbreaking approach to predict CI using convolutional neural networks (CNNs) from routine CAGs, potentially transforming clinical practices.
Methods and Analysis
The study was conducted at the Mayo Clinic, involving patients who underwent CAG and concurrent right heart catheterization from 2002 to 2023. Researchers developed a three-dimensional (3D) CNN model using the X3D architecture to predict binary CI categories—normal (CI≥2.2) versus abnormal (CI<2.2 L/min/m²)—from CAG videos. The CAGs were divided into training (70%), validation (15%), and test (15%) sets. This allowed the model to learn and validate its predictions on a diverse dataset, ensuring robust outcomes.
Results
The study included a substantial dataset of 15,297 CAG studies. Following a rigorous quality review process to exclude substandard videos, 11,475 studies involving 9,103 patients were analyzed. Among these, 25.2% displayed an abnormal CI. The model demonstrated impressive results, achieving an area under the curve (AUC) of 0.83 (95% CI 0.81 to 0.85), a sensitivity of 0.72 (95% CI 0.68 to 0.77), a specificity of 0.78 (95% CI 0.75 to 0.80), and an F1 score of 0.62 (95% CI 0.58 to 0.65) on the test dataset. Notably, using a limited selection of 4, 3, or 2 predetermined CAG videos did not significantly affect the model’s performance, indicating its flexibility and adaptability in various clinical settings.
Conclusions
This study highlights the potential of deep learning models in enhancing cardiac care. By predicting CI from CAG videos, the proposed 3D CNN model could provide real-time, non-invasive insights into a patient’s cardiac function. This advancement supports physicians in making informed decisions during CAG procedures, potentially improving patient outcomes. The integration of such technology into routine practice could revolutionize cardiac diagnostics and treatment pathways.
For further information and to explore the study in detail, please visit the source link: Here.
“`
