A Recurrent CNN for Automatic Detection and Classification of Coronary Artery Plaque and Stenosis in Coronary CT Angiography (1804.04360v4)

Abstract: Various types of atherosclerotic plaque and varying grades of stenosis could lead to different management of patients with coronary artery disease. Therefore, it is crucial to detect and classify the type of coronary artery plaque, as well as to detect and determine the degree of coronary artery stenosis. This study includes retrospectively collected clinically obtained coronary CT angiography (CCTA) scans of 163 patients. To perform automatic analysis for coronary artery plaque and stenosis classification, a multi-task recurrent convolutional neural network is applied on multi-planar reformatted (MPR) images of the coronary arteries. First, a 3D convolutional neural network is utilized to extract features along the coronary artery. Subsequently, the extracted features are aggregated by a recurrent neural network that performs two simultaneous multi-class classification tasks. In the first task, the network detects and characterizes the type of the coronary artery plaque (no plaque, non-calcified, mixed, calcified). In the second task, the network detects and determines the anatomical significance of the coronary artery stenosis (no stenosis, non-significant i.e. <50% luminal narrowing, significant i.e. >50% luminal narrowing). For detection and classification of coronary plaque, the method achieved an accuracy of 0.77. For detection and classification of stenosis, the method achieved an accuracy of 0.80. The results demonstrate that automatic detection and classification of coronary artery plaque and stenosis are feasible. This may enable automated triage of patients to those without coronary plaque and those with coronary plaque and stenosis in need for further cardiovascular workup.

• The paper introduces a dual-task RCNN that merges CNN and RNN architectures to effectively detect and classify coronary artery plaque and stenosis in CT scans.
• It utilizes multi-planar reformatted images from 163 patient scans and achieves segment-level accuracies of up to 0.80 for stenosis detection.
• The approach reduces manual evaluation variability and offers a promising step toward clinical integration for managing coronary artery disease.

Automatic Detection and Classification of Coronary Artery Plaque and Stenosis Using Recurrent CNN

The paper presents a sophisticated method utilizing a Recurrent Convolutional Neural Network (RCNN) for the automatic detection and classification of coronary artery plaques, along with the determination of the anatomical significance of stenosis in Coronary CT Angiography (CCTA) scans. With coronary artery disease (CAD) remaining a prevalent heart condition, accurate identification and classification of coronary artery plaque and stenosis are imperative for patient management and treatment strategies. The RCNN largely innovates upon existing practices by employing deep learning techniques to overcome challenges inherent in manual evaluations and traditional imaging methods.

Key Contributions and Methodology

The paper involves a dataset of retrospectively collected CCTA scans from 163 patients. The methodology employed presents several technical advancements, including:

1. Use of Multi-Planar Reformatted (MPR) Images: The coronary arteries' centerlines were extracted to form MPR images, enabling better visualization and facilitating the integration of the RCNN for feature extraction and classification.
2. Application of RCNN: A dual-task RCNN was constructed, combining the feature extraction capabilities of CNNs with the sequence analysis strengths of RNNs. This endows the network with the ability to simultaneously perform multi-class classifications pertaining to plaque type and stenosis significance.
3. Training and Evaluation: The network trained on 98 patients and tested on 65, was evaluated for segment-level, artery-level, and patient-level accuracy of classification, achieving promising results with accuracies of up to 0.80 for stenosis detection on a segment level.

The RCNN's capability of simultaneously detecting plaque type and assessing stenosis reflects a strong application of deep learning methodologies to medical image analysis, potentially reducing interobserver variability and alleviating workloads in clinical environments.

Numerical Results

The RCNN accomplished an accuracy of 0.77 for plaque detection and characterization, with a slightly higher accuracy of 0.80 for stenosis significance determination. The paper also reports a balanced approach in addressing both calcified and non-calcified plaques, simplifying manual annotation procedures with increased robustness in the automatic analysis.

Implications and Future Directions

The findings underscore the feasibility of employing deep learning models in clinical settings, particularly in the automated triage of patients based on plaque and stenosis evaluations. While the current paper demonstrates competent performance levels, potential improvements could be pursued by extending the depth of CNN layers or employing more extensive datasets with higher diversity. Additionally, the method's generalization ability across datasets from different vendors remains a pivotal question for future clinical applicability.

The proposed RCNN opens avenues for further research exploring multi-class classification tasks in medical imaging, potentially optimizing early detection systems and improving cardiac care workflows. Beyond stenosis and plaque detection, future advances may explore comprehensive coronary assessments encompassing functional significance, such as fractional flow reserve measurements, further integrating the benefits of non-invasive CCTA with advanced machine learning solutions.

Overall, this paper contributes significantly to the dialogue on automated medical diagnostics, demonstrating an effective model suitable for integration into clinical setups, pending broader validation.