Artificial intelligence-enabled plaque characterization from coronary computed tomography establishes basis of angina in women with nonobstructive atherosclerosis Authors: Rafal Wolny, Guadalupe Flores Tomasino, Kajetan Grodecki, Joel Lenell, Caroline Park, Rebekah Park, Jacek Kwiecinski, Osama Dasa, Vinicius Calsavara, Daniel S Berman, Piotr J Slomka, Matthew J Budoff, Eileen Handberg, Carl J Pepine, Leslee J Shaw, Janet Wei, Martha Gulati, C Noel Bairey Merz, Balaji Tamarappoo, Damini Dey Published: 2026 ABSTRACT AbstractBackgroundHalf of women with ischemic symptoms have non-obstructive coronary artery disease (CAD), while the pathophysiology of their condition has not been characterized. Noncalcified (NCP) and low-attenuation plaque (CT density<30 Hounsfield units, LAP) burden quantified from coronary computed tomography angiography (CCTA) is associated with ischemia in patients with obstructive CAD. We hypothesize that NCP burden is related to angina in women with ischemic symptoms and Non-Obstructive Coronary Arteries (INOCA).MethodsWomen with INOCA enrolled in the WARRIOR trial were evaluated for angina severity with Seattle Angina Questionnaire (SAQ) at study entry. Baseline CCTA of 117 women were quantitatively analyzed with AI-based software for NCP, LAP and calcified plaque (CP) volumes and burdens (%, normalized to vessel volume) across the coronary tree. Machine-learning ischemia risk score (ML-IRS) integrating quantitative lumen and plaque features from CCTA was automatically measured.ResultsAmong 109 women with visible plaque on CCTA (age 61.9, SD 10.3 years) median total plaque burden is 26.6% (IQR 18.6,32.0) and median SAQ score is 61.4 (IQR 54.6,69.1). Patients with more severe angina (SAQ ≤ 60) are younger (58.1 vs 62.0 years, p = 0.015), have higher total cholesterol (195 vs 165 mg/dL, p = 0.006), but less frequently receive statins (31.8 vs 64.4%, p = 0.006) compared with patients with SAQ > 60. Patients with SAQ ≤ 60 have higher total plaque (33.3 vs 24.3%, p = 0.001), and NCP burden (33.3 vs. 23.2%, p = 0.00065), and lower CP burden (0.0 vs. 0.3%, p = 0.005) compared with patients with SAQ > 60. On multivariable linear regression adjusted for risk factors, higher NCP burden (β = −0.50, p = 0.001), LAP burden (β = −4.50, p = 0.008) and ML-IRS (β = −3.09, p = 0.04) are associated with lower SAQ score, i.e. more severe angina.ConclusionsIn women with INOCA, high-risk atherosclerotic plaque phenotypes are related to more severe angina. Subject terms: Vascular diseases, Chronic pain Introduction Approximately half of women with ischemic symptoms which include angina have no obstructive coronary artery disease (INOCA) on anatomical imaging1,2. Yet, these women have higher than expected risk for major adverse cardiac events vs age and sex-matched asymptomatic subjects and have a high prevalence of coronary microvascular dysfunction2–5. Stenosis severity and plaque characteristics are well established determinants of myocardial ischemia in obstructive coronary artery disease (CAD). However, whether plaque characteristics are associated with angina severity in women with INOCA has not been reported. Anatomical imaging of coronary vasculature with coronary computed tomography angiography (CCTA) recently became a standard of care for evaluation of patients with stable chest pain and has significant value in cardiovascular risk assessment6,7. Beyond defining stenosis severity, artificial intelligence (AI)-based automated plaque assessment by CCTA provides a comprehensive quantitative characterization of atherosclerotic plaque composition. Several studies have shown that adverse plaque characteristics quantified from CCTA including noncalcified plaque (NCP) and low attenuation noncalcified plaque (LAP) burden are associated with an increased risk of myocardial ischemia and adverse cardiac events8,9. Accordingly, we hypothesize that adverse coronary plaque characteristics, specifically NCP and LAP burden, may be associated with ischemic symptoms in women with suspected INOCA. For the present study we utilize baseline clinical and imaging data of a subgroup of patients enrolled in the WARRIOR trial. The results show a relationship between CCTA-derived high-risk coronary plaque burden with severity of angina in women with INOCA. Thus we provide insights into the pathophysiology of symptoms in INOCA. Methods Population The Women’s IschemiA TRial to Reduce Events In Non-ObstRuctive CAD (WARRIOR NCT03417388, first registered on 24th Jan 2018) trial is a multicenter, prospective, randomized, blinded outcome evaluation of intensive medical therapy (with aspirin, high intensity statins and angiotensin converting enzyme inhibitors or angiotensin receptor blockers) vs. usual care in women with INOCA determined by invasive angiography or CCTA. WARRIOR Ancillary Study for CCTA Analysis comprised women who fulfilled the following: 1) first CCTA scan site in Los Angeles or at the University of Florida; 2) Good image quality of the first scan with present plaque; 3) Adherence to IMT if randomized to this arm10. Women enrolled in the WARRIOR Ancillary Study with nonobstructive CAD defined as luminal diameter stenosis <50% on baseline CCTA who underwent SAQ assessment were included in the current analysis. The study was approved by the institutional review boards at Cedars Sinai Medical Center, Los Angeles, CA. and University of Florida, Gainesville, FL. The authors had permission to publish the results of this independent mechanistic, National Institutes of Health-funded ancillary study independently of the main endpoints of the WARRIOR trial. All participants gave written informed consent. Coronary CCTA acquisition CCTA was performed in accordance with the Society of Cardiovascular CT guidelines11. When needed, oral and/or intravenous metoprolol was administered to achieve a target heart rate of ≤70 beats/min. Immediately before CCTA scanning, 0.4 or 0.8 mg of sublingual nitroglycerin was administered to ensure optimal coronary dilatation. Images were acquired after a bolus injection of 80-100 ml contrast medium (Omnipaque or Visipaque, GE Healthcare, Princeton, NJ) at a rate of 5-6 ml/s. The region from the carina to the diaphragm was imaged using both systolic and diastolic prospective acquisition or dose-modulated helical scanning. Acquired datasets were reconstructed at 0.6-mm slice thickness, 0.3-mm slice increment, 250-mm field-of-view, and 512 × 512 matrix. Coronary plaque analysis The coronary artery tree was evaluated using AI-enabled plaque quantification software (Autoplaque (APQ) version 3.0, Cedars-Sinai Medical Center, Los Angeles, CA, USA) with fine corrections as needed by experienced core laboratory readers12. Interobserver reproducibility was previously shown to be high for both semi-automated and AI-enabled vessel segmentation and plaque quantification performed with Autoplaque13,14. Also the age and sex-specific thresholds of automated AI-assisted plaque volumes for prediction of myocardial infarction were recently validated in a large external cohort of patients with stable CAD12. We provide the results of the interobserver variability testing for the current study in the Supplemental Table 1. Per-vessel plaque volumes (mm3) were calculated for the following components: total plaque (TP), calcified plaque (CP), noncalcified plaque (NCP), and low attenuation plaque (LAP; <30 Hounsfield units). (Fig. 1). The respective plaque burdens (%) were calculated as: plaque volume/ analyzed vessel volume×100. Scan-specific thresholds for NCP and CP were automatically generated. For each lesion, maximum diameter stenosis (DS) and contrast density difference (CDD) were calculated. DS was calculated as the ratio between the narrowest lumen diameter and the mean of two healthy, non-diseased adjacent reference points. CDD was calculated as the maximum percent difference in luminal contrast densities with respect to the proximal reference cross section without disease. Maximum CDD per patient was reported. AI-enabled plaque characterization from coronary computed tomography angiography over the coronary tree. A Curved multiplanar reformat of the left anterior descending artery. B Three dimensional volume-rendered reconstruction of the entire coronary tree with quantitative plaque overlay. Red overlay: noncalcified plaque; yellow overlay: calcified plaque. Seattle angina questionnaire The Seattle Angina Questionnaire (SAQ), was developed to assess symptoms of angina in patients with CAD. It consists of 19 items across five dimensions of symptoms, including physical limitation (PL), angina stability (AS), angina frequency (AF) treatment satisfaction (TS) and quality of life (QOL)15. SAQ scores are expressed on a 0–100 range for each domain and higher scores indicate less angina burden. A change of five or more points in the SAQ score is typically perceived as a change by the patient15. Each patient completed the SAQ at the study entry, within a maximum of 3 months form the CCTA acquisition. Machine learning ischemia risk score The previously described and validated machine learning ischemia risk score (ML-IRS) was automatically computed by the APQ software16. ML-IRS integrates quantitative CCTA measures, including NCP and LAP burden, CDD, maximum DS and plaque length. The output of the original ML-IRS are continuous values from 0 to 1 with higher value indicating greater ischemic probability. A SD-normalized ML-IRS score was used in the current analysis. Statistical analysis Continuous variables are expressed as means with standard deviations for normally distributed variables and median with interquartile ranges (IQR) for non-normally distributed variables. Normality was assessed with the Shapiro-Wilk test. Comparisons of continuous variables were performed using Wilcoxon’s rank sum test and Student’s t test if the normality assumption is satisfied. Categorical variables were expressed as counts with percentages and compared using χ2 test or Fisher’s exact test. Spearman’s statistic was applied to evaluate the correlations between clinical and imaging variables with the SAQ. Univariable and multivariable linear regression was applied to assess the association of clinical and imaging parameters, as well as ML-IRS with SAQ score. This was a pre-specified analysis according to the NIH grant 1R01HL151266-01. Covariates included in the final model were only significant predictors of SAQ by univariable linear regression and selected based on clinical relevance and prior literature. Model selection was carried out using a stepwise variable selection procedure on the basis of Akaike Information Criterion17. Model assumptions were assessed using diagnostic plots. Normality of residuals was examined using Q–Q plots. The possibility of multicollinearity was assessed by the variance inflation factor. Results from the univariable and multivariable linear regression model are presented as estimated regression coefficients (β) along with their 95% confidence intervals and p-values. All hypotheses were two-sided and tested at a 5% significance level. Statistical analyses were performed using R package version 4.4.118. Results Clinical characteristics Of the 117 enrolled women, eight were excluded because they had no apparent atherosclerosis on CCTA, thus a total of 109 women were included in the final analysis (Table 1 summarizes pertinent baseline characteristics). Their age was 61.9 years (SD 10.3), total cholesterol was 173.7 mg/dL (SD 38.3) and 63 (57.8%) of participants were receiving statins prior to randomization. Median SAQ was 73.8 (IQR 63.7, 83.9) suggesting infrequent angina. Median total coronary plaque volume was 135.6 mm3 (IQR 68.9, 245.6) and median total plaque burden was 26.3% (IQR 18.6, 31.9%). Table 1Clinical characteristicsOverall (N = 109)SAQ < 60 (N = 22)SAQ ≥ 60 (N = 87)P-valueAge, y61.4 (54.6, 69.1)58.1 (52.1, 61.8)62.0 (55.3, 69.9)0.015Body mass index, kg/m228.8 (24.7, 34.1)29.9 (27.2, 35.7)28.2 (24.3, 33.1)0.13Current smoking, n (%)6 (5.6)3 (13.6)3 (3.5)0.08Family history of CAD, n (%)82 (75.2)19 (86.4)63 (72.4)0.18Hyperlipidemia, n (%)60 (55.1)8 (36.4)52 (59.8)0.049Hypertension, n (%)55 (50.5)11 (50.0)44 (50.6)0.96Diabetes, n (%)19 (17.4)4 (18.2)15 (17.2)0.99Total cholesterol, mg/dL175.0 (146.0, 200.3)195.0 (175.8, 207.5)165.00 (138.8, 195.8)0.006LDL-cholesterol, mg/dL93.0 (72.0, 116.0)107.0 (100.0, 123.0)87.0 (67.0, 110.0)0.003HDL-cholesterol, mg/dL58.5 (49.0, 68.3)56.5 (48.3, 66.8)59.00 (50.0, 68.8)0.89Triglycerides, mg/dL88.5 (70.8, 134.5)109.00 (79.5, 144.8)85.00 (68.0, 119.8)0.036Glucose, mg/dL93.0 (87.0, 104.0)90.00 (78.5, 99.5)93.00 (88.0, 109.0)0.48Creatinine, mg/dL0.77 (0.69, 0.84)0.72 (0.69, 0.81)0.78 (0.69, 0.83)0.90Systolic blood pressure, mmHg125.0 (115.8, 135.5)122.8 (111.9, 131.6)126.00 (117.3, 136.0)0.30Diastolic blood pressure, mmHg76.7 (68.7, 83.5)77.0 (69.7, 83.5)76.3 (68.3, 83.3)0.99Heart rate, beats/min68.5 (62.6, 76.4)68.0 (62.0, 73.0)68.5 (63.0, 76.5)0.62Statin use, n (%)63 (57.8)7 (31.8)56 (64.4)0.006Angiotensin converting enzyme inhibitor use, n(%)22 (20.2)6 (27.3)16 (18.4)0.38Angiotensin receptor blocker use, n(%)23 (21.1)4 (18.2)19 (21.8)0.90SAQ score61.4 (54.6, 69.1)SAQ Seattle angina questionnaire, CAD coronary artery disease, LDL low-density lipoprotein, HDL high-density lipoprotein; Values are medians with interquartile ranges or counts with percentages. Wilcoxon rank sum test was used for continuous variables; Fisher’s exact test or Pearson’s Chi-squared test were used for categorical variables. All tests are two-sided. Clinical characteristics SAQ Seattle angina questionnaire, CAD coronary artery disease, LDL low-density lipoprotein, HDL high-density lipoprotein; Values are medians with interquartile ranges or counts with percentages. Wilcoxon rank sum test was used for continuous variables; Fisher’s exact test or Pearson’s Chi-squared test were used for categorical variables. All tests are two-sided. Comparison of patients with more and less severe symptoms Patients were dichotomized based on the SAQ score to those with weekly angina, SAQ ≤ 60 (n = 22) and less frequent angina, SAQ > 60 (n = 87). Women with SAQ ≤ 60 were younger (58.1 years [IQR 52.1, 61.8] vs. 62.0 years [IQR 55.3, 70.0], p = 0.015); fewer of them had hyperlipidemia (8 [36.4%] vs. 52 [59.8%]) and less frequently were taking statin (7 [31.8%] vs. 56 [64.4%], p = 0.006) when compared with women with less frequent angina. They also had higher total cholesterol, LDL and triglycerides concentrations (Table 1). On quantitative CCTA (Table 2), women with SAQ ≤ 60 had greater total plaque burden compared to women with SAQ > 60 (33.3% [IQR 25.2, 35.2] vs. 24.3% [IQR 17.4, 30.7], p = 0.001). Plaque in these women was predominantly composed of NCP when compared with women with less frequent angina (33.6% [25.2, 35.2] vs. 21.7% [IQR 13.1, 26.6], p = 0.00065). CP burden was similarly increased in women with SAQ ≤ 60 compared with patients with SAQ > 60 (0.0% [IQR 0.00, 0.00] vs. 0.3% [IQR 0.0, 1.86], p = 0.005). LAP burden was higher in women with SAQ ≤ 60 compared with SAQ > 60 (1.0% [IQR 0.4, 1.3] vs. 0.7% [IQR 0.2, 1.1] p = 0.19 respectively). Table 2CCTA-derived quantitative plaque characteristicsOverallSAQ < 60SAQ ≥ 60p-valueTotal plaque burden, %26.6 (18.6, 32.0)33.3 (25.2, 35.2)24.3 (17.4, 30.7)0.001Noncalcified plaque burden%25.1 (16.9, 31.1)33.3 (25.2, 35.2)23.15 (14.7, 29.3)0.00065Low-density noncalcified plaque burden, %0.76 (0.30, 1.20)0.98 (0.41, 1.33)0.66 (0.20, 1.12)0.19Calcified plaque burden, %0.14 (0.00, 1.39)0.00 (0.00, 0.00)0.30 (0.00, 1.86)0.005Maximum diameter stenosis, %31.7 (20.7, 43.8)37.1 (20.8, 48.3)31.0 (20.7, 43.3)0.38Total plaque length, mm42.9 (21.5, 75.8)51.3 (28.4, 75.3)35.2 (21.4, 75.3)0.53Contrast density difference, %22.5 (11.2, 29.9)23.0 (10.1, 30.0)21.9 (11.5, 29.9)0.79ML-IRS1.1 (0.8, 1.8)1.2 (0.9, 1.7)1.0 (0.8, 1.8)0.45Values are medians (IQR). CCTA coronary computed tomography angiography, SAQ Seattle angina questionnaire, ML-IRS machine-learning ischemia risk score. Wilcoxon rank sum test was used for continuous variables; Fisher’s exact test or Pearson’s Chi-squared test were used for categorical variables. All tests are two-sided. CCTA-derived quantitative plaque characteristics Values are medians (IQR). CCTA coronary computed tomography angiography, SAQ Seattle angina questionnaire, ML-IRS machine-learning ischemia risk score. Wilcoxon rank sum test was used for continuous variables; Fisher’s exact test or Pearson’s Chi-squared test were used for categorical variables. All tests are two-sided. CCTA-derived standardized ML-IRS was higher in patients with SAQ ≤ 60 compared with patients with SAQ > 60 (1.6 [IQR 0.9, 2.4] vs. 1.1 [IQR 0.8, 1.7] p = 0.035 respectively). The maximum DS, total plaque length and maximum CDD were not significantly different between the two groups. Prediction of angina using regression analyses In univariable linear regression analysis, clinical covariates associated with lower SAQ score were characterized by younger age, higher BMI, no prior hyperlipidemia, no statin treatment, higher LDL-cholesterol and total cholesterol concentration, greater total, NCP and LAP burden and lower CP burden. (Table 3) Table 3Univariable linear regression for prediction of SAQCovariateBeta coefficient (95% CI)p-valueAge, years0.463 (0.181 to 0.745)0.002BMI, 1 kg/m2-0.458 (-0.863 to -0.052)0.029History of hyperlipidemia10.2 (4.43 to 16.0)0.00078LDL cholesterol, mg/dL-0.167 (-0.271 to -0.063)0.002Total cholesterol, mg/dL-0.114 (-0.194 to -0.035)0.006Statin use10.4 (4.57 to 16.2)0.00069Total plaque burden, %-0.536 (-0.776 to -0.295)0.00003Noncalcified plaque burden, %-0.577 (-0.811 to -0.343)0.000005Low density noncalcified plaque burden, %-5.49 (-8.86 to -2.12)0.002Calcified plaque burden, %1.94 (0.416 to 3.46)0.014ML-IRS (SD)-3.06 (-6.01 to -0.099)0.045BMI body mass index, LDL low-density lipoprotein, ML-IRS machine-learning ischemia risk score, SD standard deviation. Univariable linear regression for prediction of SAQ BMI body mass index, LDL low-density lipoprotein, ML-IRS machine-learning ischemia risk score, SD standard deviation. In a multivariable linear model, an increase in NCP and LAP burden were significantly associated with lower SAQ (Tables 4 and 5), whereas calcified plaque burden was not associated with SAQ (Table 6). When ML-IRS which incorporates clinical and CCTA-derived variables including high-risk plaque characteristics instead of individual plaque components, it was independently associated with weekly angina (SAQ ≤ 60), along with higher total cholesterol concentration (Table 7). Table 4Multivarable linear regression model for prediction of SAQ by clinical and plaque covariates (total and noncalcified component)CovariateModel 1Model 2*Beta (95% CI)p-valueBeta (95% CI)p-valueTotal plaque burden, %0.483 (−1.099 to 2.064)0.551Noncalcified plaque burden, %−0.962 (−2.548 to 0.625)0.238−0.496 (−0.724 to −0.268)0.000045Age, years0.154 (−0.181 to 0.489)0.3710.245 (−0.049 to 0.539)0.106BMI, kg/m2−0.097 (−0.536 to 0.343)0.667Hyperlipidemia5.382 (−1.481 to 12.245)0.1287.239 (1.408 to 13.070)0.017Statin use2.987 (−4.216 to 10.190)0.418R²0.2880.277Adjusted R²0.2450.256BMI body mass index.* Based on stepwise (backward) selection method using AIC criterion.CI confidence interval.Variance inflation factor (VIF) for noncalcified plaque burden, age and hyperlipidemia is 1.043, 1.194 and 1.180 respectively. Multivarable linear regression model for prediction of SAQ by clinical and plaque covariates (total and noncalcified component) BMI body mass index. * Based on stepwise (backward) selection method using AIC criterion.CI confidence interval. Variance inflation factor (VIF) for noncalcified plaque burden, age and hyperlipidemia is 1.043, 1.194 and 1.180 respectively. Table 5Multivarable linear regression model for prediction of SAQ by clinical and plaque covariates (low-density noncalcified component)CovariateModel 1Model 2*Beta (95% CI)p-valueBeta (95% CI)p-valueLow density noncalcified plaque burden, %−4.442 (−7.957 to −0.927)0.015−4.500 (−7.766 to −1.235)0.008Age, years0.210 (−0.127 to 0.548)0.2250.254 (−0.057 to 0.564)0.113BMI, kg/m20.022 (−0.466 to 0.510)0.929Hyperlipidemia6.310 (−0.768 to 13.389)0.0848.237 (2.146 to 14.329)0.009Statin use4.040 (−3.447 to 11.527)0.293R²0.2140.205Adjusted R²0.1740.181BMI body mass index.* Based on stepwise (backward) selection method using AIC criterion.CI confidence intervalVIF for low-density noncalcified plaque burden, age and hyperlipidemia is 1.045, 1.214 and 1.171 respectively. Multivarable linear regression model for prediction of SAQ by clinical and plaque covariates (low-density noncalcified component) BMI body mass index. * Based on stepwise (backward) selection method using AIC criterion.CI confidence interval VIF for low-density noncalcified plaque burden, age and hyperlipidemia is 1.045, 1.214 and 1.171 respectively. Table 6Multivarable linear regression model for prediction of SAQ by clinical and plaque covariates (calcified plaque component)CovariateModel 1Model 2*Beta (95% CI)p-valueBeta (95% CI)p-valueCalcified plaque burden, %0.649 (-1.048 to 2.346)0.455Age, years0.195 (-0.165 to 0.554)0.2910.334 (0.020 to 0.649)0.040BMI, kg/m2-0.194 (-0.663 to 0.275)0.419Hyperlipidemia5.970 (-1.395 to 13.335)0.1158.343 (2.068 to 14.618)0.011Statin use3.613 (-4.117 to 11.342)0.362R²0.1700.148Adjusted R²0.1290.131BMI body mass index.* Based on stepwise (backward) selection method using AIC criterion.CI confidence intervalVIF for age and hyperlipidemia is 1.171. Multivarable linear regression model for prediction of SAQ by clinical and plaque covariates (calcified plaque component) BMI body mass index. * Based on stepwise (backward) selection method using AIC criterion.CI confidence interval VIF for age and hyperlipidemia is 1.171. Table 7Multivarable linear regression model for prediction of SAQ by clinical covariates and ML-IRSCovariateBeta Coefficient (95% CI)p-valueAge, years0.166 (−0.188 to 0.521)0.360BMI, kg/m2−0.227 (−0.680 to 0.226)0.328Hyperlipidemia7.979 (1.561 to 14.397)0.017Total cholesterol, mg/dL−0.090 (−0.168 to −0.012)0.026ML-IRS (SD)−3.094 (−6.000 to −0.189)0.040BMI body mass index, ML-IRS machine-learning ischemia risk score, SD standard deviation. Multivarable linear regression model for prediction of SAQ by clinical covariates and ML-IRS BMI body mass index, ML-IRS machine-learning ischemia risk score, SD standard deviation. Discussion In this prospective cohort of women with INOCA enrolled in a multicenter trial, we observed using CCTA that: 1) women with more frequent angina have a greater total plaque burden, defined by CCTA, compared to those with milder symptoms; 2) this difference is mainly driven by a greater burden of high-risk components (NCP and LAP) and 3) a machine-learning risk score combining quantitative CCTA with clinical variables is independently associated with symptom severity. Symptoms of CAD in women are often different compared to men, leading to underdiagnosis and undertreatment of ischemic heart disease1. In women symptoms may be disproportionate to stenosis severity. We and others have observed that 40–65% of women with ischemic symptoms do not have obstructive CAD defined as epicardial luminal stenosis of >50% diameter reduction2,19. Women with obstructive CAD enrolled in the ISCHEMIA trial had more frequent angina despite less extensive CAD on anatomical imaging compared with men20,21. Furthermore, symptomatic women are at greater risk of adverse cardiac outcomes when matched with asymptomatic women with similar extent of CAD22. Thus, angina symptoms may be attributed to factors independent of stenosis severity especially in women with INOCA. The pathophysiology of INOCA in women is particularly complex and incompletely understood. Microvascular and endothelial dysfunction, reduced vasodilatory capacity, increased vascular resistance as well as specific plaque morphology may be important contributors23. CCTA is increasingly used as the initial noninvasive diagnostic test to exclude obstructive epicardial CAD in patients with ischemic symptoms and offers a unique opportunity to comprehensively evaluate both the lumen and the vessel wall24. With automated deep-learning, CCTA can provide a granular quantitative assessment of plaque composition with volumes and burdens of CP, NCP, and LAP. Per-patient quantitative NCP and LAP volumes have been shown to independently predict cardiac mortality and acute coronary syndromes8,9. Although women presenting with angina have less atherosclerotic plaque of all subtypes compared with men, LAP burden predicts adverse outcomes irrespective of sex25. Moreover, CCTA-derived non-plaque imaging biomarkers of perivascular inflammation were shown to improve cardiovascular risk prediction in patients with INOCA26,27. It had been presumed that myocardial ischemia is primarily related to severity of epicardial artery stenosis28. However, current evidence suggests significant association of coronary plaque morphology with ischemia irrespective of stenosis severity. The ISCHEMIA trial demonstrated that 13% of participants had INOCA, despite having documented moderate or severe ischemia29. Our group provided the first report on the association of qualitatively-defined high-risk plaque features (LAP and positive remodeling) on CCTA with greater inducible myocardial hypoperfusion measured by myocardial perfusion imaging30. Advances in CCTA-derived plaque quantification have paved the way in establishing a relationship between NCP burden and impaired myocardial flow reserve on 13N-ammonia positron emission tomography independent of stenosis severity31. This finding was validated in a study using ischemia evaluation by invasive fractional flow reserve32. Recently in a subgroup of 211 patients with suspected CAD enrolled in the PACIFIC trial, LAP burden was independently associated with impaired myocardial blood flow assessed with [15O]H2O positron emission tomography33. In the same cohort a novel machine-learning algorithm based on CCTA-derived quantitative plaque features was shown to be comparable to CT-fractional flow reserve in predicting hyperemic blood flow measured with both invasive and non-invasive methods34. The association between high-risk plaque features and ischemia has been demonstrated using invasive angiography and intravascular imaging. In a series of 382 lesions presence of a thin-cap fibroatheroma on optical coherence tomography was associated with more severe ischemia quantified by invasive fractional flow reserve35. Similarly, in the PROSPECT II trial of patients with recent myocardial infarction, coronary lesions with positive fractional flow reserve were more likely to possess high-risk vulnerable plaque features detected by near-infrared intravascular ultrasound compared with nonphysiologically significant stenoses36. In another study such high-risk lesions were predictive of increased microvascular dysfunction37. Although these trials predominantly included patients with obstructive CAD, a similar association between high-risk plaque features and ischemia holds true in patients with non-obstructive CAD. In a study of 120 patients with suspected CAD who underwent both CCTA and invasive angiography with physiological assessment, qualitative high-risk plaque features were independent predictors of an impaired fractional flow reserve38. Our prior work and another recent study both have shown that angiographic measures of epicardial coronary atheroma burden predict microvascular dysfunction in women with INOCA39,40. Moreover, impaired myocardial blood flow was directly associated with more severe angina as measured by the SAQ in INOCA patients41. Unlike prior mixed-sex trials like ISCHEMIA, which focused mostly on ischemic patients with obstructive CAD, our study indicates that quantitatively assessed high-risk plaque, specifically noncalcified and low attenuation plaque burden, may contribute to anginal symptom severity6,29. To our knowledge this observation and has not been previously described in patients with CAD, especially women with INOCA. Finally, we observed that patients with higher LDL and total cholesterol concentration reported more severe anginal symptoms while patients treated with statins had less angina. There is already extensive in-vitro and in-vivo evidence of a detrimental effect of plasma cholesterol on endothelial function, mainly mediated by suppression of Kir channels42. Prior clinical research indicated a beneficial effect of statins on endothelial function in patients with INOCA, which translated into improved microvascular circulation43,44. Our results provide real-life clinical validation of these data suggesting that higher cholesterol levels and no prior statin use are associated with the development of overt atherosclerosis. Potential confounding factors could be lack of treatment adherence as well as microvascular dysfunction leading to anginal symptoms in women with INOCA. Another feature of our study is the use of a ML-IRS which integrates clinical characteristics and CCTA-derived plaque features, for prediction of symptom severity in women with INOCA. The ML-IRS was originally developed and validated in a population of patients with obstructive CAD who underwent invasive angiography. The ischemia risk score was able to distinguish between patients with normal vs an abnormal flow reserve11. We have integrated ML-IRS assessment with our quantitative CCTA analysis software to automatically compute the ischemia risk score following each analysis. Application of this tool provides insight into the symptomatic significance of nonobstructive stenoses in women in whom the clinical impact of mild atherosclerosis might be underappreciated by diagnostic testing. High ML-IRS in a patient without obstructive coronary lesions might be a signal that patient-reported symptoms could likely have a cardiac origin. Subsequently it may prompt the use of more intense anti-atherosclerotic interventions both non-pharmacological and pharmacological in women with INOCA. There are some limitations that should be mentioned. Our findings are based on a subjective patient-reported angina severity score; however, the SAQ is validated and widely accepted as a very useful survey instrument for assessment of angina in research and clinical practice. The overall angina severity in the analyzed population is moderate. Although ischemic symptoms are considered to be attributed to CAD, some contribution of non-cardiac causes of reported symptoms cannot be excluded. The ML-IRS is assumed to serve as a surrogate of myocardial ischemia. Our study does not include objective measurements of ischemia such as fractional flow reserve or myocardial metabolic measurements using invasive and/or non-invasive methods to corroborate risk score predictions. Neither invasive nor non-invasive microvascular function testing was performed which warrants future studies using multimodality imaging using positron emission tomography or cardiac magnetic resonance. Since this is a substudy of a large trial, some degree of selection bias could not be excluded. Lastly, the data on anti-anginal medication is not available. Given the above constraints, caution should be exercised when interpreting and extrapolating our results to other populations. Conclusions In women with suspected INOCA, greater total plaque burden as well as high-risk atherosclerotic plaque phenotypes characterized by increased NCP and LAP burden as demonstrated by CCTA were related to more frequent angina. These results may provide mechanistic understanding of anginal symptoms in INOCA, as well as potential treatment targets with anti-atherosclerotic strategies. Supplementary information Supplementary Information Acknowledgements This study was primarily supported by grants from the NHLBI R01HL151266 and 1R01HL148787. It was also supported by CDMRP-DoD W81XWH-17-2-0030 and McJunkin Family Foundation through funds distributed by the University of Florida, Department of Medicine, Clinical Research Consortium CDRN-1501-26692, the Edythe L. Broad and the Constance Austin Women’s Heart Research Fellowships, Cedars-Sinai Medical Center, Los Angeles, CA, the Barbra Streisand Women’s Cardiovascular Research and Education Program, Cedars-Sinai Medical Center, Los Angeles, the Linda Joy Pollin Women’s Heart Health Program, the Erika Glazer Women’s Heart Health Project, Cedars-Sinai Medical Center, Los Angeles, CA; and the VA Women’s Health Practice-Based Research Network VA HSR&D SDR 10-012. R.W. is supported by a Polish National Science Center (grant number 2021/43/D/NZ5/02434) and Medical Research Agency (grant number KPOD.07.07-IW.07-0150/24). R.P. was supported by a grant from the Miriam and Sheldon G Adelson Medical Research Foundation. Author contributions D.D. and R.W. developed the idea and designed the framework for analyses; R.W., G.F.T., C.P., K.G., J.L. and R.P. performed the CCTA-derived plaque quantification; G.F.T., C.P., O.D. and R.P. extracted clinical data; B.T., R.W., and J.L. made the figures; V.C. performed statistical analyses; R.W., D.D., B.T., J.K., D.S.B, L.S., M.B., E.H., C.J.P., J.W. and M.G. wrote and repeatedly revised the paper, D.D. supervised the project. Peer review Peer review information Communications Medicine thanks Botond Barna Mátyás, Neng Dai and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Data availability All the results supporting the findings of this study are available within the paper and in Supplementary Table 1. Deidentified final results are available for research purposes upon reasonable written request to the corresponding author. In line with the WARRIOR ancillary study protocol, access to such data requires a Data Access Agreement, which is subject to consideration by the study steering committee, the central investigational review board and the office of research administration at the Cedars-Sinai Medical Center. Competing interests Drs Berman, Slomka, and Dey have received software royalties from Cedars-Sinai Medical Center and report equity in APQ Health. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Balaji Tamarappoo, Damini Dey. Supplementary information The online version contains supplementary material available at 10.1038/s43856-026-01668-6. References Associated Data Supplementary Materials Supplementary Information Data Availability Statement All the results supporting the findings of this study are available within the paper and in Supplementary Table 1. Deidentified final results are available for research purposes upon reasonable written request to the corresponding author. In line with the WARRIOR ancillary study protocol, access to such data requires a Data Access Agreement, which is subject to consideration by the study steering committee, the central investigational review board and the office of research administration at the Cedars-Sinai Medical Center.