3D CT-Based Coronary Calcium Assessment: A Feature-Driven Machine Learning Framework

2025-10-29

AI Summary

This study addresses the challenge of scarce expert annotations for coronary artery calcium (CAC) scoring in non-contrast coronary CT angiography (CCTA). We propose a pseudo-labeling method that bypasses manual segmentation, integrating handcrafted radiomic features to build an automated CAC assessment model. Unlike mainstream approaches relying on pretrained deep models (e.g., CT-FM, RadImageNet) for feature extraction, we systematically validate and demonstrate that radiomic features significantly outperform deep features in zero/non-zero CAC classification (p < 0.05). Evaluated on a clinical dataset of 182 cases, our method achieves 84% classification accuracy, markedly enhancing early coronary artery disease risk stratification from non-enhanced scans. Our key contributions are: (1) the first integration of pseudo-labeling with radiomics—eliminating dependence on pixel-level annotations; and (2) empirical evidence establishing the distinct advantage of radiomics over deep learning in low-data CAC quantification.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Segmentation
• Reasoning under Uncertainty: Causality

Application Category

• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Coronary artery calcium (CAC) scoring plays a crucial role in the early detection and risk stratification of coronary artery disease (CAD). In this study, we focus on non-contrast coronary computed tomography angiography (CCTA) scans, which are commonly used for early calcification detection in clinical settings. To address the challenge of limited annotated data, we propose a radiomics-based pipeline that leverages pseudo-labeling to generate training labels, thereby eliminating the need for expert-defined segmentations. Additionally, we explore the use of pretrained foundation models, specifically CT-FM and RadImageNet, to extract image features, which are then used with traditional classifiers. We compare the performance of these deep learning features with that of radiomics features. Evaluation is conducted on a clinical CCTA dataset comprising 182 patients, where individuals are classified into two groups: zero versus non-zero calcium scores. We further investigate the impact of training on non-contrast datasets versus combined contrast and non-contrast datasets, with testing performed only on non contrast scans. Results show that radiomics-based models significantly outperform CNN-derived embeddings from foundation models (achieving 84% accuracy and p<0.05), despite the unavailability of expert annotations.

Problem

• Automating coronary calcium scoring without expert annotations
• Comparing radiomics features with deep learning embeddings
• Classifying patients into zero versus non-zero calcium groups

Innovation

• Radiomics-based pipeline with pseudo-labeling for training
• Pretrained foundation models extract image features for classification
• Combined contrast and non-contrast datasets enhance training