Juhwan Lee, Ph.D.
Assistant Professor, Department of Biomedical Engineering VCU College of Engineering
- Richmond VA
Dr. Juhwan Lee’s research focuses on artificial intelligence, medical image analysis, and their applications in cardiovascular medicine.
Links
Social
Biography
Industry Expertise
Areas of Expertise
Education
Case Western Reserve University
Postdoctoral fellowship
Biomedical Engineering
2020
Dongguk University
Ph.D.
Biomedical Engineering
2015
Patents
Prediction of major adverse cardiovascular events (mace) from ai analysis of pericoronary fat in ct images
US12573036B2
2026-03-10
The present disclosure, in some embodiments, relates to a method of generating a prognosis for a patient. The method includes accessing automatically segmented pericoronary adipose tissue (PCAT) corresponding to a patient within an electronic memory. A plurality of non-confounding PCAT features are generated by measuring values of Hounsfield units for an imaging unit within the PCAT. The measured values of the Hounsfield units are predominately free of iodine confounding and artifacts. The plurality of non-confounding PCAT features are provided to a regression model. The regression model is configured to generate a prognosis for the patient using the plurality of non-confounding PCAT features
Plaque segmentation in intravascular optical coherence tomography (OCT) images using deep learning
US11710238B2
2023-07-25
Embodiments discussed herein facilitate segmentation of vascular plaque, training a deep learning model to segment vascular plaque, and/or informing clinical decision-making based on segmented vascular plaque. One example embodiment accessing vascular imaging data for a patient, wherein the vascular imaging data comprises a volume of interest; pre-process the vascular imaging data to generate pre-processed vascular imaging data; provide the pre-processed vascular imaging data to a deep learning model trained to segment a lumen and a vascular plaque; and obtain segmented vascular imaging data from the deep learning model, wherein the segmented vascular imaging data comprises a segmented lumen and a segmented vascular plaque in the volume of interest.
Research Grants
Cardiovascular risk prediction from AI analysis of coronary calcifications
NIH STTR
2025-08-15
We will create software for predicting cardiovascular health from a low-cost (no-cost) CT calcium score examination, which shows calcifications in the coronary arteries. We will use artificial intelligence to greatly improve the existing CT calcium score method, creating a method that physicians and patients can use in shared decision-making to personalize interventions.
Multi-modality evaluation of high-risk coronary atherosclerotic plaque
NIH K01
2024-09-01
We will develop methods for the non-invasive, quantitative evaluation of coronary artery disease in computed tomography angiography images. With success, our research will lead to improved detection of coronary artery disease and evaluation of its severity, paving the way for personalized treatments.
Selected Articles
Sub-Agatston coronary calcification on non-contrast cardiac CT and cardiovascular risk Get access Arrow
European Journal of Preventive CardiologyPrerna Singh, Ammar Hoori, Juhwan Lee, Michelle C Williams, Sadeer Al-Kindi, David E Newby, Sanjay Rajagopalan, David L Wilson
2026-06-01
Aims
To develop assessments of low-density (‘sub-Agatston’) coronary artery calcification (CAC) on non-contrast CT calcium score (CTCS) examinations to evaluate its association with the risk of atherosclerotic major adverse cardiac events (MACE) and with high-risk plaque (HRP) features from coronary CT angiography (CCTA).
Methods and results
We analysed 1837 asymptomatic individuals from the Community Benefit of No-charge Calcium Score Screening (CLARIFY) CTCS registry (NCT04075162) and 1075 patients presenting at chest-pain clinics from the Scottish COmputed Tomography of the HEART (SCOT-HEART) trial (NCT01149590). MACE comprised death, myocardial infarction, stroke, or revascularization. We developed the vulnerable calcium index (VCI), which quantifies the proportional CAC mass increase after region-growing from 130 Hounsfield Units (HU) (Agatston) to 110 HU. Cox models were adjusted for age, sex, diabetes, hypertension, smoking, baseline statin use, and Agatston score. The association between VCI and CCTA-defined HRP (low-attenuation plaque, positive remodelling, spotty calcifications, mixed plaque, punctate calcification, napkin-ring sign) was assessed in patients with paired CTCS and CCTA in SCOT-HEART. Median follow-up was 4.8 years (CLARIFY) and 4.9 years (SCOT-HEART). VCI was independently associated with MACE in both cohorts: CLARIFY (adjusted hazard ratio [HR] 1.12, 95% confidence interval [CI] 1.02–1.22, P = 0.02) per unit increase in VCI and SCOT-HEART (adjusted HR 1.16, 1.03–1.31, P = 0.02). Patients in the top quartile of adjusted model risk had markedly higher event rates compared with the lowest quartile (CLARIFY HR 25.0; SCOT-HEART HR 33.7 vs. Q1, P < 0.005 for both). Addition of VCI to Agatston significantly improved C-index (CLARIFY: 0.62 to 0.66, P < 0.001; SCOT-HEART: 0.72 to 0.75, P < 0.001) and net reclassification index (CLARIFY: 0.52 [0.40, 0.64], P < 0.005; SCOT-HEART: 0.72 [0.48, 0.94], P < 0.005). In SCOT-HEART, VCI was associated with all HRP features (odds ratio: 1.45–1.85, P < 0.001), adjusted for Agatston score.
Conclusion
VCI captures sub-Agatston CAC linked to HRP and independently improves MACE prediction in both a primary-prevention and a mixed-risk symptomatic chest pain population. Incorporating VCI into CAC reporting may refine preventive risk stratification beyond the conventional Agatston score.
Hybrid deep learning time-to-event modeling of major adverse cardiovascular events using coronary artery calcium score scans
European Journal of Radiology Artificla IntelligenceJustin N Kim, Juhwan Lee, Ammar Hoori, Hao Wu, Tao Hu, Sadeer Al-Kindi, Mohamed Makhlouf, Robert Gilkeson, Sanjay Rajagopalan, David L Wilson
2026-09-07
Background
The Agatston score from non-contrast coronary calcium scoring scan (CACS) is a robust predictor of major adverse cardiovascular events (MACE) but summarizes total calcium burden and may miss other risk-related imaging information in CACS. We hypothesized that a hybrid deep-learning (DL) model integrating image features with clinical, Agatston, and calcium-omics data could improve MACE risk prediction.
Methods
In this retrospective cohort of 1950 patients, a DenseNet-121 encoder extracted DL-derived features from CACS volume, which were fused with tabular data (clinical features, Agatston, and 15 calcium-omics features) within a deep survival model. Performance was assessed via five-fold cross-validation using Harrell’s concordance-index (C-index) and time-dependent area under the curve (AUC), and net reclassification index (NRI), relative to tabular-only baseline.
Results
Expanding the tabular feature set from clinical variables alone to include Agatston and calcium-omics improved discrimination for both tabular-only baseline and hybrid models. Incorporating DL-derived imaging representations consistently outperformed tabular baselines in C-indices (0.802 vs. 0.789 with the full feature set) and yielded higher time-dependent AUCs during first three years of follow-up. The hybrid model also achieved an NRI of 0.119 (p = 0.037). Patients reclassified from low to high risk had a significantly higher event rate (37.5%) compared to those remaining low risk (1.1%, p
Pericoronary adipose tissue feature analysis in computed tomography calcium score images in comparison to coronary computed tomography angiography
Journal of Medical ImagingYingnan Song, Hao Wu, Juhwan Lee, Justin Kim, Ammar Hoori, Tao Hu, Vladislav Zimin, Mohamed Makhlouf, Sadeer Al-Kindi, Sanjay Rajagopalan, Chun-Ho Yun, Chung-Lieh Hung, David L Wilson
2025-01-12
Purpose: We investigated the feasibility and advantages of using non-contrast CT calcium score (CTCS) images to assess pericoronary adipose tissue (PCAT) and its association with major adverse cardiovascular events (MACE). PCAT features from coronary computed tomography angiography (CCTA) have been shown to be associated with cardiovascular risk but are potentially confounded by iodine. If PCAT in CTCS images can be similarly analyzed, it would avoid this issue and enable its inclusion in formal risk assessment from readily available, low-cost CTCS images.
Approach: To identify coronaries in CTCS images that have subtle visual evidence of vessels, we registered CTCS with paired CCTA images having coronary labels. We developed an "axial-disk" method giving regions for analyzing PCAT features in three main coronary arteries. We analyzed hand-crafted and radiomic features using univariate and multivariate logistic regression prediction of MACE and compared results against those from CCTA.
Results: Registration accuracy was sufficient to enable the identification of PCAT regions in CTCS images. Motion or beam hardening artifacts were often prevalent in "high-contrast" CCTA but not CTCS. Mean HU and volume were increased in both CTCS and CCTA for the MACE group. There were significant positive correlations between some CTCS and CCTA features, suggesting that similar characteristics were obtained. Using hand-crafted/radiomics from CTCS and CCTA, AUCs were 0.83/0.79 and 0.83/0.77, respectively, whereas Agatston gave AUC = 0.73.
Conclusions: Preliminarily, PCAT features can be assessed from three main coronary arteries in non-contrast CTCS images with performance characteristics that are at the very least comparable to CCTA.
Cardiac CT Perfusion Imaging of Pericoronary Adipose Tissue (PCAT) Highlighting Potential Confounds in CTA Analysis
Journal of Clinical MedicineHao Wu, Yingnan Song, Ammar Hoori, Juhwan Lee, Sadeer Al-Kindi, Wei-Ming Huang, Chun-Ho Yun, Chung-Lieh Hung, Sanjay Rajagopalan, David L Wilson
2025-01-24
Background: Features of pericoronary adipose tissue (PCAT) from coronary computed tomography angiography (CCTA) are associated with inflammation and cardiovascular risk. As PCAT is vascularly connected with coronary vasculature, the presence of iodine is a potential confounding factor on PCAT HU and textures that has not been adequately investigated. We aim to use dynamic cardiac CT perfusion (CCTP) to understand the perfusion of PCAT and determine its effects on PCAT assessment. Methods: From CCTP, we analyzed HU dynamics of territory-specific PCAT, the myocardium, and other adipose depots in patients with coronary artery disease. HU, blood flow, and radiomics were assessed over time. Changes from peak aorta time, Pa, chosen to model the acquisition time of CCTA, were obtained. Results: HU in PCAT increased more than in other adipose depots. Blood flow in PCAT was ~23% of that in the contiguous myocardium. A two-second offset [before, after] Pa resulted in [4 ± 1.1 HU, 3 ± 1.5 HU] differences in PCAT, giving a 7 HU swing. Due to changes in HU, the apparent PCAT volume reduced by ~15% from the first scan (P1) to Pa using a conventional fat window. Comparing radiomic features over time, 78% of features changed >10% relative to P1. Distal and proximal to a significant stenosis, we found less enhancement and longer time-to-peak distally in PCAT. Conclusions: CCTP elucidates blood flow in PCAT and enables the analysis of PCAT features over time. PCAT assessments (HU, apparent volume, and radiomics) are sensitive to acquisition timing and obstructive stenosis, which may confound the interpretation of PCAT in CCTA images. Data normalization may be in order.
Improving coronary artery segmentation with self-supervised learning and automated pericoronary adipose tissue segmentation
Journal of Medical ImagingJustin N Kim, Yingnan Song, Hao Wu, Ananya Subramaniam, Jihye Lee, Mohamed H E Kakhlouf, Neda S Hassani, Sadeer Al-Kindi, David L Wilson, Juhwan Lee
2025-01-16
Purpose: Coronary artery disease (CAD) is a leading cause of morbidity and mortality worldwide, with coronary computed tomography angiography (CCTA) playing a crucial role in its diagnosis. The mean Hounsfield unit (HU) of pericoronary adipose tissue (PCAT) is linked to cardiovascular risk. We utilized a self-supervised learning framework (SSL) to improve the accuracy and generalizability of coronary artery segmentation on CCTA volumes while addressing the limitations of small-annotated datasets.
Approach: We utilized self-supervised pretraining followed by supervised fine-tuning to segment coronary arteries. To evaluate the data efficiency of SSL, we varied the number of CCTA volumes used during pretraining. In addition, we developed an automated PCAT segmentation algorithm utilizing centerline extraction, spatial-geometric coronary identification, and landmark detection. We evaluated our method on a multi-institutional dataset by assessing coronary artery and PCAT segmentation accuracy via Dice scores and comparing mean PCAT HU values with the ground truth.
Results: Our approach significantly improved coronary artery segmentation, achieving Dice scores up to 0.787 after self-supervised pretraining. The automated PCAT segmentation achieved near-perfect performance, with
R
-squared values of 0.9998 for both the left anterior descending artery and the right coronary artery indicating excellent agreement between predicted and actual mean PCAT HU values. Self-supervised pretraining notably enhanced model generalizability on external datasets, improving overall segmentation accuracy.
Conclusions: We demonstrate the potential of SSL to advance CCTA image analysis, enabling more accurate CAD diagnostics. Our findings highlight the robustness of SSL for automated coronary artery and PCAT segmentation, offering promising advancements in cardiovascular care.
Prediction of obstructive coronary artery disease using coronary calcification and epicardial adipose tissue assessments from CT calcium scoring scans
Journal of Cardiovascular Computed TomographyJuhwan Lee, Tao Hu, Michelle C. Williams, Ammar Hoori, Hao Wu, Justin N. Kim, David E. Newby, Robert Gilkeson, Sanjay Rajagopalan, David L. Wilson
2025-04-16
Background
Low-cost/no-cost non-contrast CT calcium scoring (CTCS) exams can provide direct evidence of coronary atherosclerosis. In this study, using features from CTCS images, we developed a novel machine learning model to predict obstructive coronary artery disease (CAD), as defined by the coronary artery disease-reporting and data system (CAD-RADS).
Methods
This study analyzed 1324 patients from the SCOT-HEART trial who underwent both CTCS and CT angiography. Obstructive CAD was defined as CAD-RADS 4A-5, while CAD-RADS 0–3 were considered non-obstructive CAD. We analyzed clinical, Agatston-score-derived, and epicardial fat-omics features to predict obstructive CAD. The most predictive features were selected using elastic net logistic regression and used to train a CatBoost model. Model performance was evaluated using 1000 repeated five-fold cross-validation and survival analyses to predict major adverse cardiovascular event (MACE) and revascularization. Generalizability was assessed using an external validation set of 2316 patients for survival predictions.
Results
Among the 1324 patients, obstructive CAD was identified in 334 patients (25.2 %). Elastic net regression identified the top 14 features (5 clinical, 2 Agatston-score-derived, and 7 fat-omics). The proposed method achieved excellent performance for classifying obstructive CAD, with an AUC of 90.1 ± 0.9 % and sensitivity/specificity/accuracy of 83.5 ± 5.5 %/93.7 ± 1.9 %/82.4 ± 2.0 %. The inclusion of Agatston-score-derived and fat-omics features significantly improved classification performance. Survival analyses showed that both actual and predicted obstructive CAD significantly differentiated patients who experienced MACE and revascularization.
Conclusions
We developed a novel machine learning model to predict obstructive CAD from non-contrast CTCS scans. Our findings highlight the potential clinical benefits of CTCS imaging in identifying patients likely to benefit from advanced imaging.
Detection of arterial remodeling using epicardial adipose tissue assessment from CT calcium scoring scan
Frontiers in Cardiovascular MedicineJuhwan Lee, Tao Hu, Michelle C. Williams, Ammar Hoori, Hao Wu, Justin N. Kim, David E. Newby, Robert Gilkeson, Sanjay Rajagopalan, David L. Wilson
2025-03-13
Introduction: Non-contrast CT calcium scoring (CTCS) exams have been widely used to assess coronary artery disease. However, their clinical applications in predicting coronary arterial remodeling remain unknown. This study aimed to develop a novel machine learning model to predict positive remodeling (PR) from CTCS scans and evaluate its clinical value in predicting major adverse cardiovascular events (MACE).Methods: We analyzed data from 1,324 patients who underwent both CTCS and CT angiography. PR was defined as an outer vessel diameter at least 10% greater than the average diameter of the segments immediately proximal and distal to the plaque. We utilized a total of 246 features, including 23 clinical features, 12 Agatston score-derived features, and 211 epicardial fat-omics features to predict PR. Feature selection was performed using Elastic Net logistic regression, and the selected features were used to train a CatBoost machine learning model. Classification performance was evaluated using 1,000 repetitions of five-fold cross-validation and survival analyses, comparing actual and predicted PR in the context of predicting MACE.Results: PR was identified in 429 patients (32.4%). Using Elastic Net, we identified the top 13 features, including four clinical features, three Agatston score-derived features, and six fat-omics features. Our method demonstrated excellent classification performance for predicting PR, achieving a sensitivity of 80.3 ± 1.7%, a specificity of 89.7 ± 1.7%, and accuracy of 81.9 ± 2.5%. The Agatston-score-derived and fat-omics features provided additional benefits, improving classification performance. Furthermore, our model effectively predicted MACE, with a hazard ratio (HR) of 4.5 [95% confidence interval (CI): 3.2–6.4; C-index: 0.578; p
Computational analysis of intravascular OCT images for future clinical support: a comprehensive review
IEEE Reviews in Biomedical EngineeringJuhwan Lee, Yazan Gharaibeh, Pengfei Dong, Luis A P Dallan, Gabriel T R Pereira, Justin N Kim, Ammar Hoori, Linxia Gu, Hiram G Bezerra, Bernardo Cortese, David L Wilson
2025-01-16
Over the past two decades, intravascular optical coherence tomography (IVOCT) has emerged as a promising tool for planning percutaneous coronary interventions (PCI), studying coronary artery disease, and assessing treatments. With its near-histological resolution and optical contrast, IVOCT uniquely evaluates coronary plaque characteristics, enhancing the guidance of interventional procedures. Artificial intelligence (AI) techniques have been widely applied to IVOCT imaging, providing fast and accurate automated interpretation. These techniques hold significant potential for both clinical and research purposes. Clinically, automated analysis offers comprehensive assessments of coronary plaques, leading to better treatment decisions during PCI. For research, automated interpretation of IVOCT opens new avenues to understand the pathophysiology of coronary atherosclerosis. However, these techniques face several limitations, including issues related to spatial resolution, challenges in manual assessments, and the additional time required for these analyses. This review covers recent advancements and applications of AI techniques and computational simulation methods in IVOCT image analysis, including vessel wall segmentation, plaque characterization, stent analysis, and their clinical applications. Furthermore, we discuss the potential of AI-enhanced IVOCT analysis to facilitate personalized decision-making, potentially improving short- and long-term patient outcomes.
