Translating Perivascular Inflammation Imaging into Precision Coronary Care: An Ex plainable Nomogram Integrating Pericoronary Adipose Tissue Attenuation Index with Anatomical and Functional CT Biomarkers for Acute Coronary Syndrome Discrimination

Abstract Background: Acute coronary syndrome (ACS) and stable coronary artery disease (SCAD) present overlapping clinical phenotypes, yet mandate divergent therapeutic strategies—urgent revascularization versus pharmacological optimization. Current diagnostic reliance on high-sensitivity cardiac troponin dynamic monitoring introduces hazardous delays (4–6 hours detection window) and yields false negatives in 20% of non-ST-elevation myocardial infarction. Pericoronary adipose tissue attenuation index (FAI), quantifying vascular inflammation through computed tomography (CT) attenuation changes in pericoronary fat, offers mechanistic insight into coronary plaque destabilization. However, the translational pathway from this biomarker to clinically actionable decision support remains incompletely realized, particularly regarding integration with established anatomical and functional parameters and explicit clinical interpretability. Hypothesis: A mechanistically grounded, multi-dimensional nomogram combining FAI with coronary CT angiography (CCTA)-derived anatomical stenosis (%DS), left ventricular ejection fraction (LVEF), and smoking status—rendered interpretable through SHapley Additive exPlanations (SHAP)—would achieve robust ACS-SCAD discrimination while providing transparent, auditable decision logic suitable for clinical implementation. Methods: In this prospective diagnostic accuracy study adhering to TRIPOD 2019 guidelines, 388 patients (ACS 117 [30.2%], SCAD 271 [69.8%]) underwent CCTA with standardized FAI, %DS, and LVEF acquisition. Stratified randomization (7:3) produced training (n=271) and validation (n=117) sets with preserved outcome distribution. Least absolute shrinkage and selection operator (LASSO) regression with 10-fold stratified cross-validation compared lambda.min versus lambda.1se selection criteria, followed by multivariable logistic regression. Performance was assessed by area under the receiver operating characteristic curve (AUC), calibration, Brier score, and decision curve analysis. KernelSHAP parsed individualized decision contributions. Regulatory-readiness was evaluated through predetermined change control documentation and explanation stability assessment. Results: LASSO lambda.min retained %DS, FAI, LVEF, RVEF, and smoking; lambda.1se selected only %DS and FAI. Final model incorporated %DS, FAI, LVEF, and smoking based on mechanistic non-overlap (RVEF-LVEF functional redundancy) and epidemiological importance. Multivariable analysis demonstrated %DS (OR 1.03, 95% CI: 1.01–1.04, P < 0.001) and FAI (OR 1.04, 95% CI: 1.02–1.06, P < 0.001) as positive ACS predictors; LVEF as protective (OR 0.96, 95% CI: 0.92–0.99, P = 0.032); smoking as marginally non-significant (OR 1.83, 95% CI: 0.87–3.86, P = 0.110) but incrementally contributory. Nomogram achieved AUC 0.86 (95% CI: 0.82–0.91) in training and 0.82 (95% CI: 0.75–0.90) in validation. Optimal cutoff 0.291 (Youden index) yielded training sensitivity 0.80, specificity 0.82, negative predictive value 0.91; validation sensitivity 0.73, specificity 0.72, negative predictive value 0.83. Calibration was satisfactory (Hosmer-Lemeshow P = 0.347 and P = 0.405). Decision curve analysis demonstrated net benefit across threshold probability 0.10–0.55. SHAP analysis established a mechanistic hierarchy: %DS and FAI as dominant drivers (mean |SHAP| 0.75 and 0.70), LVEF as a functional protective modifier (0.35), and smoking as a behavioral risk stratifier (0.20). Individual force plots enabled case-level explanation with documented contribution stability. Conclusions: The %DS-FAI-LVEF-smoking nomogram achieves favorable ACS discriminatory accuracy with acceptable generalization. SHAP-enhanced interpretability satisfies emerging transparency requirements for artificial intelligence-enabled medical devices, supporting regulatory-ready clinical translation. Translational Significance: This study bridges the critical T1–T2 translational gap by converting perivascular inflammation biology (FAI) into an evidence-based diagnostic tool validated for a specific clinical decision context. The "one-stop" CCTA acquisition strategy—yielding anatomical, inflammatory, and functional data from a single examination—addresses implementation barriers in time-critical emergency settings. Explicit SHAP-based explainability exceeds current "black box" machine learning limitations, aligning with FDA and EMA guidance on predetermined change control and transparency documentation for clinical artificial intelligence. These design elements collectively reduce translational friction from research prototype to prospective clinical trials and eventual regulatory approval, advancing precision coronary care through mechanism-informed, transparently interpretable decision support.

Authors

Publication Details

Journal
Research Square
Published
2026-11-14
DOI
https://doi.org/10.21203/rs.3.rs-11139418/v1
Primary Topic
Cardiovascular Disease and Adiposity
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Translating Perivascular Inflammation Imaging into Precision Coronary Care: An Ex plainable Nomogram Integrating Pericoronary Adipose Tissue Attenuation Index with Anatomical and Functional CT Biomarkers for Acute Coronary Syndrome Discrimination

JinJun Dong, YinLong Qi, Sheng Zhou, Ting Ni et al.
3 links
Research Square
Cardiovascular Disease and Adiposity
preprint

Translating Perivascular Inflammation Imaging into Precision Coronary Care: An Ex plainable Nomogram Integrating Pericoronary Adipose Tissue Attenuation Index with Anatomical and Functional CT Biomarkers for Acute Coronary Syndrome Discrimination

JinJun Dong, YinLong Qi, Sheng Zhou, Ting Ni, Mingzhu Yu, Xin Liu, Haiyang Zhang, YaNan Li
preprint en

Abstract

Abstract Background: Acute coronary syndrome (ACS) and stable coronary artery disease (SCAD) present overlapping clinical phenotypes, yet mandate divergent therapeutic strategies—urgent revascularization versus pharmacological optimization. Current diagnostic reliance on high-sensitivity cardiac troponin dynamic monitoring introduces hazardous delays (4–6 hours detection window) and yields false negatives in 20% of non-ST-elevation myocardial infarction. Pericoronary adipose tissue attenuation index (FAI), quantifying vascular inflammation through computed tomography (CT) attenuation changes in pericoronary fat, offers mechanistic insight into coronary plaque destabilization. However, the translational pathway from this biomarker to clinically actionable decision support remains incompletely realized, particularly regarding integration with established anatomical and functional parameters and explicit clinical interpretability. Hypothesis: A mechanistically grounded, multi-dimensional nomogram combining FAI with coronary CT angiography (CCTA)-derived anatomical stenosis (%DS), left ventricular ejection fraction (LVEF), and smoking status—rendered interpretable through SHapley Additive exPlanations (SHAP)—would achieve robust ACS-SCAD discrimination while providing transparent, auditable decision logic suitable for clinical implementation. Methods: In this prospective diagnostic accuracy study adhering to TRIPOD 2019 guidelines, 388 patients (ACS 117 [30.2%], SCAD 271 [69.8%]) underwent CCTA with standardized FAI, %DS, and LVEF acquisition. Stratified randomization (7:3) produced training (n=271) and validation (n=117) sets with preserved outcome distribution. Least absolute shrinkage and selection operator (LASSO) regression with 10-fold stratified cross-validation compared lambda.min versus lambda.1se selection criteria, followed by multivariable logistic regression. Performance was assessed by area under the receiver operating characteristic curve (AUC), calibration, Brier score, and decision curve analysis. KernelSHAP parsed individualized decision contributions. Regulatory-readiness was evaluated through predetermined change control documentation and explanation stability assessment. Results: LASSO lambda.min retained %DS, FAI, LVEF, RVEF, and smoking; lambda.1se selected only %DS and FAI. Final model incorporated %DS, FAI, LVEF, and smoking based on mechanistic non-overlap (RVEF-LVEF functional redundancy) and epidemiological importance. Multivariable analysis demonstrated %DS (OR 1.03, 95% CI: 1.01–1.04, P < 0.001) and FAI (OR 1.04, 95% CI: 1.02–1.06, P < 0.001) as positive ACS predictors; LVEF as protective (OR 0.96, 95% CI: 0.92–0.99, P = 0.032); smoking as marginally non-significant (OR 1.83, 95% CI: 0.87–3.86, P = 0.110) but incrementally contributory. Nomogram achieved AUC 0.86 (95% CI: 0.82–0.91) in training and 0.82 (95% CI: 0.75–0.90) in validation. Optimal cutoff 0.291 (Youden index) yielded training sensitivity 0.80, specificity 0.82, negative predictive value 0.91; validation sensitivity 0.73, specificity 0.72, negative predictive value 0.83. Calibration was satisfactory (Hosmer-Lemeshow P = 0.347 and P = 0.405). Decision curve analysis demonstrated net benefit across threshold probability 0.10–0.55. SHAP analysis established a mechanistic hierarchy: %DS and FAI as dominant drivers (mean |SHAP| 0.75 and 0.70), LVEF as a functional protective modifier (0.35), and smoking as a behavioral risk stratifier (0.20). Individual force plots enabled case-level explanation with documented contribution stability. Conclusions: The %DS-FAI-LVEF-smoking nomogram achieves favorable ACS discriminatory accuracy with acceptable generalization. SHAP-enhanced interpretability satisfies emerging transparency requirements for artificial intelligence-enabled medical devices, supporting regulatory-ready clinical translation. Translational Significance: This study bridges the critical T1–T2 translational gap by converting perivascular inflammation biology (FAI) into an evidence-based diagnostic tool validated for a specific clinical decision context. The "one-stop" CCTA acquisition strategy—yielding anatomical, inflammatory, and functional data from a single examination—addresses implementation barriers in time-critical emergency settings. Explicit SHAP-based explainability exceeds current "black box" machine learning limitations, aligning with FDA and EMA guidance on predetermined change control and transparency documentation for clinical artificial intelligence. These design elements collectively reduce translational friction from research prototype to prospective clinical trials and eventual regulatory approval, advancing precision coronary care through mechanism-informed, transparently interpretable decision support.

Research Square
National Health Commission of the People's Republic of China
Good health and well-being
Cardiovascular Disease and Adiposity
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon