Advances in Diagnosis, Therapy, and Prognosis of Coronary Artery Disease Powered by Deep Learning Algorithms

Figures

• 

  Figure 1. The Pipeline of Developing a Deep Learning Model The pipeline includes training, validation, and testing phases using independent nonoverlapping data sets. The training phase optimizes model parameters by minimizing the cost function, the validation phase tunes the hyperparameters related to training strategy, and the testing phase evaluates the performance of the final model. The overlapping of the data used in the 3 phases should be strictly avoided. CAD = coronary artery disease.

• 

  Figure 2. The Development of DL Model Architecture in Chronological Order Milestones in the development of deep learning (DL) models include the feedforward neural network (FNN), recurrent neural network (RNN), convolutional neural network (CNN), graph neural network (GNN), generative adversarial network (GAN), and Transformer.

• 

  Figure 3. Typical Computational Tasks of DL Models With Corresponding Evaluation Metrics Typical computational tasks include classification, segmentation, detection, reconstruction, and regression. CCTA = coronary computed tomography angiography; CMRA = coronary magnetic resonance angiography; DCGAN = deep convolutional generative adversarial network; DL = deep learning; GAN = generative adversarial network; IVUS = intravascular ultrasound; OCT = optical coherence tomography; RCNN = region-based convolutional neural network; VGG = visual geometry group; XA = X-ray angiography.

• 

  Central Illustration. Clinical Applications of Deep Learning in the Workflow of Interventional Catheterization Deep learning models open new opportunities to improve multiple aspects of the interventional workflow, involving image acquisition, diagnosis and risk stratification, interventional navigation, and therapy evaluation and optimization. AI = artificial intelligence.