Integrated Pipeline for Coronary Angiography With Automated Lesion Profiling, Virtual Stenting, and 100-Vessel FFR Validation
Summary
AngioAI-QFR introduces an end-to-end angiography-only pipeline leveraging deep learning for automated coronary artery disease assessment, integrating lesion profiling, virtual stenting, and automated angiography-derived Fractional Flow Reserve (QFR). The system demonstrates strong correlation with invasive FFR (r=0.89, MAE=0.045) and high diagnostic accuracy (AUC 0.93) for detecting hemodynamically significant lesions, operating near real-time. This practical pipeline unifies computer vision, functional profiling, and virtual PCI planning.
Medical Relevance
This pipeline offers a non-invasive, wire-free method for physiologically assessing coronary artery lesions, significantly improving diagnostic accuracy and guiding percutaneous coronary intervention (PCI) planning by predicting outcomes, potentially reducing the need for invasive FFR and optimizing patient management.
AI Health Application
The AngioAI-QFR system uses deep learning and computer vision for automated stenosis detection, lumen segmentation, and virtual stenting. It provides an AI-driven, wire-free physiological assessment (quantitative flow ratio - QFR) for coronary artery disease, aiming to improve diagnostic accuracy and facilitate treatment planning for percutaneous coronary intervention (PCI) in a near real-time clinical workflow.
Key Points
- Addresses the limitations of visual grading variability in coronary angiography and the workflow intensity of existing wire-based FFR and angiography-derived QFR tools.
- The AngioAI-QFR pipeline integrates deep learning for stenosis detection, lumen segmentation, centreline/diameter extraction, per-millimetre Relative Flow Capacity (RFC) profiling, and virtual stenting.
- It automatically recomputes angiography-derived Quantitative Flow Ratio (QFR) after virtual stenting, facilitating pre-procedural planning for percutaneous coronary intervention (PCI).
- Evaluated against invasive FFR as the reference standard in 100 consecutive vessels, AngioAI-QFR showed a strong correlation (r = 0.89, MAE = 0.045) with FFR.
- Achieved high diagnostic performance for detecting FFR <= 0.80, with an Area Under the Curve (AUC) of 0.93, sensitivity of 0.88, and specificity of 0.86.
- The system demonstrated high automation, completing fully automatically in 93% of vessels, with a rapid median processing time of 41 seconds.
- RFC profiling distinguished between focal and diffuse patterns of capacity loss, and virtual stenting predicted larger QFR gains in focal disease, enhancing lesion characterization and treatment guidance.
Methodology
The researchers developed AngioAI-QFR, an end-to-end angiography-only pipeline. It utilizes deep learning for automated stenosis detection and accurate lumen segmentation, followed by centreline and diameter extraction. The system then computes per-millimetre Relative Flow Capacity (RFC) profiles and angiography-derived Quantitative Flow Ratio (QFR). A virtual stenting module allows for simulated stent placement with automatic recomputation of QFR. The pipeline was validated in 100 consecutive vessels against invasive wire-based FFR as the gold standard, assessing correlation, mean absolute error, and diagnostic performance.
Key Findings
Deep learning models achieved high precision (0.97) for stenosis detection and Dice coefficient (0.78) for lumen segmentation. AngioAI-QFR demonstrated strong agreement with invasive FFR (r = 0.89, MAE = 0.045) across 100 vessels. Its diagnostic performance for FFR <= 0.80 was excellent (AUC 0.93, sensitivity 0.88, specificity 0.86). The pipeline operated mostly automatically (93% of vessels) with a rapid median result time of 41 seconds. RFC profiling successfully differentiated focal from diffuse capacity loss, and virtual stenting accurately predicted larger QFR gains in focal lesions, aiding treatment strategy.
Clinical Impact
AngioAI-QFR provides a practical, near real-time tool for non-invasive physiological assessment of coronary lesions, streamlining diagnosis and treatment planning during catheterization. It can enhance lesion selection for PCI, facilitate virtual stenting to optimize procedure strategy and predict outcomes, potentially reducing the need for invasive FFR and improving patient care by guiding more precise and effective interventions.
Limitations
The abstract does not explicitly state specific limitations or caveats regarding the AngioAI-QFR pipeline or the study's design (e.g., specific patient cohort, generalizability, long-term outcomes, or detailed comparison against all existing QFR solutions). It primarily highlights the limitations of current practices which this solution aims to overcome.
Future Directions
The abstract does not explicitly mention future research directions. It presents the AngioAI-QFR pipeline as a developed and validated solution poised for practical application in clinical settings.
Medical Domains
Keywords
Abstract
Coronary angiography is the main tool for assessing coronary artery disease, but visual grading of stenosis is variable and only moderately related to ischaemia. Wire based fractional flow reserve (FFR) improves lesion selection but is not used systematically. Angiography derived indices such as quantitative flow ratio (QFR) offer wire free physiology, yet many tools are workflow intensive and separate from automated anatomy analysis and virtual PCI planning. We developed AngioAI-QFR, an end to end angiography only pipeline combining deep learning stenosis detection, lumen segmentation, centreline and diameter extraction, per millimetre Relative Flow Capacity profiling, and virtual stenting with automatic recomputation of angiography derived QFR. The system was evaluated in 100 consecutive vessels with invasive FFR as reference. Primary endpoints were agreement with FFR (correlation, mean absolute error) and diagnostic performance for FFR <= 0.80. On held out frames, stenosis detection achieved precision 0.97 and lumen segmentation Dice 0.78. Across 100 vessels, AngioAI-QFR correlated strongly with FFR (r = 0.89, MAE 0.045). The AUC for detecting FFR <= 0.80 was 0.93, with sensitivity 0.88 and specificity 0.86. The pipeline completed fully automatically in 93 percent of vessels, with median time to result 41 s. RFC profiling distinguished focal from diffuse capacity loss, and virtual stenting predicted larger QFR gain in focal than in diffuse disease. AngioAI-QFR provides a practical, near real time pipeline that unifies computer vision, functional profiling, and virtual PCI with automated angiography derived physiology.
Comments
22 pages, 10 figures, 7 tables