Abstract:Functionally significant coronary artery disease (CAD) is caused by plaque buildup in the coronary arteries, potentially leading to narrowing of the arterial lumen, i.e. coronary stenosis, that significantly obstructs blood flow to the myocardium. The current reference for establishing the presence of a functionally significant stenosis is invasive fractional flow reserve (FFR) measurement. To avoid invasive measurements, non-invasive prediction of FFR from coronary CT angiography (CCTA) has emerged. For this, machine learning approaches, characterized by fast inference, are increasingly developed. However, these methods predict a single FFR value per artery i.e. they don't provide information about the stenosis location or treatment strategy. We propose a deep learning-based method to predict the FFR along the artery from CCTA scans. This study includes CCTA images of 110 patients who underwent invasive FFR pullback measurement in 112 arteries. First, a multi planar reconstruction (MPR) of the artery is fed to a variational autoencoder to characterize the artery, i.e. through the lumen area and unsupervised artery encodings. Thereafter, a convolutional neural network (CNN) predicts the FFR along the artery. The CNN is supervised by multiple loss functions, notably a loss function inspired by the Earth Mover's Distance (EMD) to predict the correct location of FFR drops and a histogram-based loss to explicitly supervise the slope of the FFR curve. To train and evaluate our model, eight-fold cross-validation was performed. The resulting FFR curves show good agreement with the reference allowing the distinction between diffuse and focal CAD distributions in most cases. Quantitative evaluation yielded a mean absolute difference in the area under the FFR pullback curve (AUPC) of 1.7. The method may pave the way towards fast, accurate, automatic prediction of FFR along the artery from CCTA.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the functional assessment of coronary artery disease (CAD). Specifically, the authors propose a deep - learning - based method, aiming to non - invasively predict the fractional flow reserve (FFR) of the entire coronary artery from coronary CT angiography (CCTA). Traditionally, functionally significant coronary artery stenoses are determined by invasive FFR measurements, which are not only costly but also burdensome for patients. However, most current machine - learning methods can only predict one FFR value for each artery and cannot provide information on the location of the stenosis or treatment strategies. Therefore, the method proposed in this paper aims to overcome these limitations and provide FFR predictions along the artery, thereby helping to locate lesions that require treatment and distinguish between diffuse and focal diseases. ### Main contributions of the paper 1. **Non - invasive prediction of FFR**: Using CCTA images, predict FFR values along the coronary artery through a deep - learning model, avoiding traditional invasive measurements. 2. **Spatial distribution information**: Unlike existing methods, this method not only predicts a single FFR value but also provides spatial distribution information of FFR values along the artery, which is helpful for accurately locating the lesion position. 3. **Automatic feature extraction**: Combine supervised and unsupervised learning to automatically extract arterial features from CCTA images, reducing the need for manual segmentation. 4. **Novel loss functions**: Propose a loss function based on the Earth Mover's Distance (EMD) and a histogram loss function to ensure the accuracy of the FFR - decreasing position and the similarity of the shape. ### Method overview 1. **Data preparation**: - Collected CCTA images of 110 patients who had undergone invasive FFR pull - back measurements. - Used multi - planar reconstruction (MPR) technology to generate high - resolution arterial images. 2. **Arterial feature extraction**: - Utilize variational auto - encoder (VAE) to extract arterial features from MPR images, including lumen area and unsupervised arterial coding. - Combine the coronary artery centerline tree to extract additional features, such as branch and collateral information. 3. **FFR prediction network**: - Use a 1D convolutional neural network (CNN) to analyze the extracted arterial features and predict FFR values along the artery. - In order to ensure the accuracy of the FFR - decreasing position, design a loss function based on EMD. - In order to distinguish between focal and diffuse diseases, introduce a histogram loss function. 4. **Evaluation**: - Train and evaluate the model through eight - fold cross - validation. - Compare the predicted FFR curves with the reference values and calculate the area under the precision - recall curve (AUPC), and the results show good consistency. ### Formulas - **Earth Mover's Distance (EMD) loss**: \[ EMD=\sum_{j = 0}^{l}\left|\sum_{i = 0}^{j}(\Delta FFR_{\text{pred}}(x_i)-\Delta FFR_{\text{ref}}(x_i))\right| \] - **Symmetric EMD loss**: \[ L_{\text{symm}}^{\text{FFR}}=L^{\text{FFR}}+\bar{L}^{\text{FFR}} \] where, \[ \bar{L}^{\text{FFR}}=\sum_{j = 0}^{l}\left|\sum_{i = 0}^{j}(\Delta FFR_{\text{pred}}(x_{l - i})-\Delta FFR_{\text{ref}}(x_{l - i}))\right| \] - **Histogram loss**: \[ \text{Histogram Loss}=\sum_{b}w_b\left|H_{\text{pred}}(b)-H_{\text{ref}}(b)\right| \] where \( H_{\text{pred}}(b) \) and \( H_{\text{ref}}(b) \) are respectively

Deep Learning-Based Prediction of Fractional Flow Reserve along the Coronary Artery

A Novel Physics-Based Model for Fast Computation of Blood Flow in Coronary Arteries

Deep learning-based detection of functionally significant stenosis in coronary CT angiography

Diagnostic Performance of Computational Fluid Dynamics (Cfd)-Based Fractional Flow Reserve (FFR) Derived from Coronary Computed Tomographic Angiography (CCTA) for Assessing Functional Severity of Coronary Lesions

A comprehensive approach to prediction of fractional flow reserve from deep-learning-augmented model

Estimation of FFR in coronary arteries with deep learning

Additional Value of Deep Learning Computed Tomographic Angiography-Based Fractional Flow Reserve in Detecting Coronary Stenosis and Predicting Outcomes.

Diagnostic Accuracy of Coronary Computed Tomography Angiography-Derived Fractional Flow Reserve

Deep learning-based coronary computed tomography analysis to predict functionally significant coronary artery stenosis

Non-invasive fractional flow reserve estimation using deep learning on intermediate left anterior descending coronary artery lesion angiography images

Diagnostic Performance of Noninvasive Coronary Computed Tomography Angiography-Derived FFR for Coronary Lesion-Specific Ischemia Based on Deep Learning Analysis

Constraint-Aware Learning for Fractional Flow Reserve Pullback Curve Estimation from Invasive Coronary Imaging

Deep learning analysis of coronary arteries in cardiac CT angiography for detection of patients requiring invasive coronary angiography

Non-invasive fractional flow reserve derived from reduced-order coronary model and machine learning prediction of stenosis flow resistance

Deep Learning-based Prediction of Coronary Artery Stenosis Resistance.

Co-registered optical coherence tomography and X-ray angiography for the prediction of fractional flow reserve

Diagnostic Performance of Angiography-Based Fractional Flow Reserve for Functional Evaluation of Coronary Artery Stenosis

Attention-based Multi-fidelity Machine Learning Model for Computational Fractional Flow Reserve Assessment

Diagnostic Performance of Angiography-Based Fractional Flow Reserve for Functional Evaluation of Coronary Artery Stenosis.

Prediction of fractional flow reserve based on reduced-order cardiovascular model

A Multi-Dimensional CFD Framework for Fast Patient-Specific Fractional Flow Reserve Prediction.