Deep learning analysis of the myocardium in coronary CT angiography for identification of patients with functionally significant coronary artery stenosis (1711.08917v2)

Abstract: In patients with coronary artery stenoses of intermediate severity, the functional significance needs to be determined. Fractional flow reserve (FFR) measurement, performed during invasive coronary angiography (ICA), is most often used in clinical practice. To reduce the number of ICA procedures, we present a method for automatic identification of patients with functionally significant coronary artery stenoses, employing deep learning analysis of the left ventricle (LV) myocardium in rest coronary CT angiography (CCTA). The study includes consecutively acquired CCTA scans of 166 patients with FFR measurements. To identify patients with a functionally significant coronary artery stenosis, analysis is performed in several stages. First, the LV myocardium is segmented using a multiscale convolutional neural network (CNN). To characterize the segmented LV myocardium, it is subsequently encoded using unsupervised convolutional autoencoder (CAE). Thereafter, patients are classified according to the presence of functionally significant stenosis using an SVM classifier based on the extracted and clustered encodings. Quantitative evaluation of LV myocardium segmentation in 20 images resulted in an average Dice coefficient of 0.91 and an average mean absolute distance between the segmented and reference LV boundaries of 0.7 mm. Classification of patients was evaluated in the remaining 126 CCTA scans in 50 10-fold cross-validation experiments and resulted in an area under the receiver operating characteristic curve of 0.74 +- 0.02. At sensitivity levels 0.60, 0.70 and 0.80, the corresponding specificity was 0.77, 0.71 and 0.59, respectively. The results demonstrate that automatic analysis of the LV myocardium in a single CCTA scan acquired at rest, without assessment of the anatomy of the coronary arteries, can be used to identify patients with functionally significant coronary artery stenosis.

• The paper presents a deep learning method using CNNs, CAEs, and SVMs to identify functionally significant stenosis by analyzing the left ventricular myocardium in CCTA scans, aiming for a non-invasive approach.
• The approach involves precise LV myocardial segmentation using a multi-scale CNN achieving a Dice coefficient of 0.91, followed by unsupervised feature extraction using a convolutional autoencoder.
• The proposed method achieved an AUC of 0.74 in classifying patients, suggesting its potential clinical use in reducing invasive procedures by analyzing myocardial characteristics rather than solely focusing on artery stenosis.

Deep Learning for Myocardial Analysis in CCTA: Identifying Functionally Significant Coronary Stenosis

The paper presented in the paper focuses on the development and validation of a deep learning-based methodology aimed at identifying patients with functionally significant coronary artery stenosis using coronary computed tomography angiography (CCTA). Traditional assessment of stenosis severity often relies on fractional flow reserve (FFR) measurements obtained during invasive coronary angiography (ICA), which can be resource-intensive and carries inherent procedural risks. This paper offers an alternative by employing an automated pipeline that analyzes the left ventricular (LV) myocardium in CCTA scans acquired at rest, thus potentially reducing the dependency on invasive procedures.

Methodological Insights

The proposed methodology involves several stages, starting with the segmentation of the LV myocardium using a multi-scale convolutional neural network (CNN). This is a critical step as precise segmentation is essential for the subsequent characterization of the myocardium. The multi-scale approach in the CNN allows the extraction of both local texture and broader contextual information. The paper reports a robust Dice coefficient of 0.91, indicating high accuracy in segmentation when compared to manual annotations.

Following segmentation, the myocardium is encoded using a convolutional autoencoder (CAE). The CAE framework enables the extraction of myocardial texture features in an unsupervised manner, capturing subtleties that may indicate ischemic changes. The model then divides the myocardium into spatially connected clusters, computing statistical measures of the encodings which are subsequently used to classify the patients using a support vector machine (SVM) classifier.

Quantitative and Qualitative Results

The classification process was evaluated using an area under the receiver operating characteristic curve (AUC) metric, achieving a score of 0.74 with a standard deviation of 0.02 in 10-fold cross-validation experiments across 126 patient scans. This outcome, while modest, suggests that the approach has a reasonable capacity to discern patients who require further interventions based on their myocardial characteristics alone. Sensitivity and specificity metrics demonstrated balanced performance, with specificities of 0.77, 0.71, and 0.59 at sensitivity levels of 0.60, 0.70, and 0.80 respectively.

Implications and Future Directions

The paper's findings have significant implications for clinical workflows by potentially reducing the number of patients subjected to invasive FFR measurements, thus minimizing procedural risks and healthcare costs. Additionally, the research highlights the utility of deep learning frameworks in processing medical imaging data, diverging from the traditional stenosis analysis approaches that focus heavily on the coronary artery tree.

The authors acknowledge limitations, including the dependence on the accuracy of LV myocardium segmentation and the absence of specific stenosis localization. Future research might delve into integrating this algorithmic approach with computational fluid dynamics models to enhance diagnostic accuracy and incorporate a more comprehensive analysis of the coronary anatomy alongside myocardial evaluation.

Moreover, expanding validation to larger datasets from diverse demographics and imaging systems would fortify the generalizability of the findings. As deep learning continues to evolve, future iterations of this method could leverage advancements in 3D analysis techniques and federated learning approaches, ensuring robustness across varying imaging conditions and patient populations.

In conclusion, this paper outlines a methodological advancement in utilizing CCTA imaging and deep learning to infer the functional significance of coronary artery stenosis, laying a foundation for future innovations aimed at non-invasive cardiac diagnostics.