The increasing prevalence of cardiovascular disease (CVD) necessitates efficient methods for early detection and risk assessment. Coronary Artery Calcium Scoring (CACS) using computed tomography (CT) has long been established as a reliable tool to identify coronary artery calcifications (CAC) and predict potential cardiovascular events. Traditionally, radiologists have manually conducted these evaluations using semi-automated software. However, the rise of artificial intelligence (AI) presents new opportunities to streamline and enhance this process. A recent study examined the effectiveness of AI-based coronary calcium scoring compared to expert semi-automatic evaluations, using data from the Swedish CArdioPulmonary bioImage Study (SCAPIS).
AI in Coronary Calcium Scoring: Promise and Potential
Artificial intelligence (AI) has emerged as a powerful tool in radiology, offering the potential to assist or even replace human experts in certain diagnostic tasks. When applied to coronary calcium scoring, AI aims to reduce the time and resources spent on manual evaluations while maintaining or enhancing accuracy. In the SCAPIS study, over 4500 calcium-scoring CT (CSCT) examinations were assessed to compare AI-derived scores with semi-automatic evaluations by expert readers. The results showed an impressive correlation between AI and expert evaluations, demonstrating the potential of AI to support clinical decision-making and alleviate the growing workload in radiology departments.
The AI software's ability to generate accurate coronary calcium scores without human intervention could transform how radiologists approach cardiovascular risk assessments. Not only does it save time, but it also ensures efficient handling of large datasets, such as those from population-wide studies. By reducing the need for manual evaluation, AI allows radiologists to focus on more complex cases, ultimately improving patient care.
Accuracy and Agreement in AI-Based Scoring
A key objective of the study was to evaluate the accuracy and agreement of AI-based calcium scoring compared to the traditional semi-automatic method. The metrics assessed included the Agatston score (AS), volume score (VS), mass score (MS), and the number and location of lesions. The findings revealed excellent correlation, with intraclass correlation coefficients (ICCs) reaching 0.994 for AS, VS, and MS. This suggests that AI can provide consistent and reliable results across various scoring metrics.
Additionally, the study explored the accuracy of AI in categorising patients into different cardiovascular risk groups based on their Agatston scores. Here, AI achieved a remarkable 91.2% accuracy, with a weighted kappa analysis indicating almost perfect agreement between AI and expert readers. These findings highlight the robustness of AI in accurately predicting cardiovascular risk and reinforce its potential for widespread clinical use.
However, it is important to note that some discrepancies occurred. In a small percentage of cases, the AI overestimated or underestimated the risk category, with 7.2% of cases being overestimated and 1.6% underestimated. Despite these minor variations, the overall performance of AI suggests it is a valuable tool for supporting radiologists, especially in large-scale evaluations where human error or time constraints may affect accuracy.
Challenges and Future Directions
While the study's results are promising, certain challenges remain. The generalisability of the AI model was limited by the fact that the study relied on data from a single CT system at Linköping University Hospital. The authors acknowledge that future studies should incorporate data from diverse CT systems and populations to confirm the robustness of AI models across different clinical settings. The lack of variation in the data acquisition process may have influenced the AI’s performance, which could differ when applied to other equipment or protocols.
Furthermore, although AI showed the potential to reduce workload and improve efficiency, the study emphasised the importance of maintaining human oversight in the diagnostic process. The AI model was prone to misregistrations, particularly when dealing with image noise or non-coronary calcifications. Human intervention was necessary in these instances to correct misclassifications and ensure accurate cardiovascular risk assessments. AI systems must be refined to minimise these errors and improve their standalone diagnostic capabilities.
The study also points to the need to continuously test AI algorithms on larger and more varied datasets. With improvements in AI technology, future models could provide even greater accuracy, with fewer discrepancies between AI and expert evaluations. Striking the right balance between automation and human expertise in radiology will be crucial.
In conclusion, the study highlights the significant promise of AI-based coronary calcium scoring in clinical practice. With excellent correlation and agreement with semi-automatic evaluations, AI presents a valuable tool for improving the efficiency and accuracy of cardiovascular risk assessments. While challenges remain, particularly in generalisability and error management, the potential for AI to transform coronary calcium scoring is clear. As AI technology advances, its role in enhancing patient care by reducing radiologists' workloads and supporting precise diagnostics will undoubtedly grow. Further research involving more diverse datasets and clinical environments will help solidify AI’s place as a trusted aid in the fight against cardiovascular disease.
Source: European Radiology
Image Credit: iStock
