Investigating the Feasibility of Using DenseNet to Improve Coronary Calcification Detection in CT

Optimized Lightweight Architecture for Coronary Artery Disease Classification in Medical Imaging

Background/Objectives: The early and accurate detection of Coronary Artery Disease (CAD) is crucial for preventing life-threatening complications, particularly among athletes engaged in high-intensity endurance sports. This demographic faces unique cardiovascular risks, as prolonged and intense physical exertion can exacerbate underlying CAD conditions. Studies indicate that while athletes typically exhibit enhanced cardiovascular health, this demographic is not immune to Coronary Artery Disease (CAD) risks. Research has shown that approximately 1-2% of competitive athletes suffer from CAD-related complications, with sudden cardiac arrest being the leading cause of mortality in athletes over 35 years old. High-intensity endurance sports can exacerbate underlying CAD conditions due to the prolonged physical stress placed on the cardiovascular system, making early detection crucial. This study aimed to develop and evaluate a lightweight deep learning model for CAD detection tailored to the unique challenges of diagnosing athletes. Methods: This study introduces a lightweight deep learning model specifically designed for CAD detection in athletes. By integrating ResNet-inspired residual connections into the VGG16 architecture, the model achieves a balance of high diagnostic accuracy and computational efficiency. By incorporating ResNet-inspired residual connections into the VGG16 architecture, the model enhances gradient flow, mitigates vanishing gradient issues, and improves feature extraction of subtle morphological variations in coronary lesions. Its lightweight design, with only 1.2 million parameters and 3.5 GFLOPs, ensures suitability for real-time deployment in resource-constrained clinical environments, such as sports clinics and mobile diagnostic systems, where rapid and efficient diagnostics are essential for high-risk populations. Results: The proposed model achieved superior performance compared to state-of-the-art architectures, with an accuracy of 90.3%, recall of 89%, precision of 90%, and an AUC-ROC of 0.912. These metrics highlight its robustness in detecting and classifying CAD in athletes. The model lightweight architecture, with only 1.2 million parameters and 3.5 GFLOPs, ensures computational efficiency and suitability for real-time clinical applications, particularly in resource-constrained settings. Conclusions: This study demonstrates the potential of a lightweight, deep learning-based diagnostic tool for CAD detection in athletes, achieving a balance of high diagnostic accuracy and computational efficiency. Future work should focus on integrating broader dataset validations and enhancing model explainability to improve adoption in real-world clinical scenarios.

A CNN-based denoising method trained with images acquired with electron density phantoms for thin-sliced coronary artery calcium scans

PURPOSE

This work proposed a convolutional neural network (CNN)-based method trained with images acquired with electron density phantoms to reduce quantum noise for coronary artery calcium (CAC) scans reconstructed with slice thickness less than 3 mm.

METHODS

A DenseNet model was used to estimate quantum noise for CAC scans reconstructed with slice thickness of 0.5, 1.0 and 1.5 mm. Training data was acquired using electron density phantoms in three different sizes. The label images of the CNN model were real noise maps, while the input images of the CNN model were pseudo noise maps. Image denoising was conducted by subtracting the CNN output images from thin-sliced CAC scans. The efficacy of the proposed method was verified through both phantom study and patient study.

RESULTS

By means of phantom study, the proposed method was proven effective in reducing quantum noise in CAC scans reconstructed with 1.5-mm slice thickness without causing significant texture change or variation in HU values. With regard to patient study, calcifications were more clear on the denoised CAC scans reconstructed with slice thickness of 0.5, 1.0 and 1.5 mm than on 3-mm slice images, while over-smooth changes were not observed in the denoised CAC scans reconstructed with 1.5-mm slice thickness.

CONCLUSION

Our results demonstrated that the electron density phantoms can be used to generate training data for the proposed CNN-based denoising method to reduce quantum noise for CAC scans reconstructed with 1.5-mm slice thickness. Because anthropomorphic phantom is not a necessity, our method could make image denoising more practical in routine clinical practice.