Implicancias de la inteligencia artificial en los métodos de imagen endovascular / Implications of Artificial Intelligence in Intravascular Imaging Methods

RESUMEN La angioplastia transluminal coronaria (ATC) es una de las principales estrategias de revascularización en pacientes con enfermedad coronaria aterosclerótica (ECA). Numerosos estudios respaldan la optimización de la ATC mediante métodos de imagen endovascular; sin embargo, estos métodos son subutilizados en la práctica clínica contemporánea y enfrentan desafíos en la interpretación de los datos obtenidos, por lo que la integración de la inteligencia artificial (IA) se vislumbra como una solución atractiva para promover y simplificar su uso. La IA se define como un programa computarizado que imita la capacidad del cerebro humano para recopilar y procesar datos. El aprendizaje de máquinas es una subdisciplina de la IA que implica la creación de algoritmos capaces de analizar grandes conjuntos de datos sin suposiciones previas, mientras que el aprendizaje profundo se centra en la construcción y entrenamiento de redes neuronales artificiales profundas y complejas. Así, se ha demostrado que la incorporación de sistemas de IA a los métodos de imagen endovascular incrementa la precisión de la ATC, disminuye el tiempo del procedimiento y la variabilidad interobservador en la interpretación de los datos obtenidos, promueve así una mayor adopción y facilita su utilización. El propósito de la presente revisión es destacar cómo los sistemas actuales basados en IA pueden desempeñar un papel fundamental en la interpretación de los datos generados por los métodos de imagen endovascular, lo que conduce a una mejora en la optimización de la ATC en pacientes con ECA.

ABSTRACT Percutaneous coronary intervention (PCI) is one of the primary revascularization strategies in patients with coronary artery disease (CAD). Several studies support the use of intravascular imaging methods to optimize PCI. However, these methods are underutilized in contemporary clinical practice and face challenges in data interpretation. Therefore, the incorporation of artificial intelligence (AI) is seen as an attractive solution to promote and simplify their use. AI can be defined as a computer program that mimics the human brain in its ability to collect and process data. Machine learning is a sub-discipline of AI that involves the creation of algorithms capable of analyzing large datasets without making prior assumptions, while deep learning focuses on the construction and training of deep and complex artificial neural networks. The incorporation of AI systems to intravascular imaging methods improves the accuracy of PCI, reduces procedure duration, and minimizes interobserver variability in data interpretation. This promotes their wider adoption and facilitates their use. The aim of this review is to highlight how current AI-based systems can play a key role in the interpretation of data generated by intravascular imaging methods and optimize PCI in patients with CAD.

Data-driven models for the prediction of coronary atherosclerotic plaque progression/regression.

Coronary artery disease is defined by the existence of atherosclerotic plaque on the arterial wall, which can cause blood flow impairment, or plaque rupture, and ultimately lead to myocardial ischemia. Intravascular ultrasound (IVUS) imaging can provide a detailed characterization of lumen and vessel features, and so plaque burden, in coronary vessels. Prediction of the regions in a vascular segment where plaque burden can either increase (progression) or decrease (regression) following a certain therapy, has remained an elusive major milestone in cardiology. Studies like IBIS-4 showed an association between plaque burden regression and high-intensity rosuvastatin therapy over 13 months. Nevertheless, it has not been possible to predict if a patient would respond in a favorable/adverse fashion to such a treatment. This work aims to (i) Develop a framework that processes lumen and vessel cross-sectional contours and extracts geometric descriptors from baseline and follow-up IVUS pullbacks; and to (ii) Develop, train, and validate a machine learning model based on baseline/follow-up IVUS datasets that predicts future percent of atheroma volume changes in coronary vascular segments using only baseline information, i.e. geometric features and clinical data. This is a post hoc analysis, revisiting the IBIS-4 study. We employed 140 arteries, from 81 patients, for which expert delineation of lumen and vessel contours were available at baseline and 13-month follow-up. Contour data from baseline and follow-up pullbacks were co-registered and then processed to extract several frame-wise features, e.g. areas, plaque burden, eccentricity, etc. Each pullback was divided into regions of interest (ROIs), following different criteria. Frame-wise features were condensed into region-wise markers using tools from statistics, signal processing, and information theory. Finally, a stratified 5-fold cross-validation strategy (20 repetitions) was used to train/validate an XGBoost regression models. A feature selection method before the model training was also applied. When the models were trained/validated on ROI defined by the difference between follow-up and baseline plaque burden, the average accuracy and Mathews correlation coefficient were 0.70 and 0.41 respectively. Using a ROI partition criterion based only on the baseline's plaque burden resulted in averages of 0.60 accuracy and 0.23 Mathews correlation coefficient. An XGBoost model was capable of predicting plaque progression/regression changes in coronary vascular segments of patients treated with rosuvastatin therapy in 13 months. The proposed method, first of its kind, successfully managed to address the problem of stratification of patients at risk of coronary plaque progression, using IVUS images and standard patient clinical data.

Comparison of endothelial shear stress between ultrathin strut bioresorbable polymer drug-eluting stent vs durable-polymer drug-eluting stent post-stent implantation: An optical coherence tomography substudy from BIOFLOW II.

BACKGROUND: Recent clinical data indicate a different performance of biodegradable polymer (BP)-drug eluting stent (DES) compared to durable polymer (DP)-DES. Whether this can be explained by a beneficial impact of BP-DES stent design on the local hemodynamic forces distribution remains unclear. OBJECTIVES: To compare endothelial shear stress (ESS) distribution after implantation of ultrathin (us) BP-DES and DP-DES and examine the association between ESS and neointimal thickness (NIT) distribution in the two devices at 9 months follow up. METHODS AND RESULTS: We retrospectively identified patients from the BIOFLOW II trial that had undergone OCT imaging. OCT data were utilized to reconstruct the surface of the stented segment at baseline and 9 months follow-up, simulate blood flow, and measure ESS and NIT in the stented segment. The patients were divided into 3 groups depending on whether DP-DES (N = 8, n = 56,160 sectors), BP-DES with a stent diameter of >3 mm (strut thickness of 80 µm, N = 6, n = 36,504 sectors), or BP-DES with a stent diameter of ≤3 mm (strut thickness of 60 µm, N = 8, n = 50,040 sectors) were used for treatment. The ESS, and NIT distribution and the association of these two variables were estimated and compared among the 3 groups. RESULTS: In the DP-DES group mean NIT was 0.18 ± 0.17 mm and ESS 1.68 ± 1.66 Pa; for the BP-DES ≤3 mm group the NIT was 0.17 ± 0.11 mm and ESS 1.49 ± 1.24 Pa and for the BP-DES >3 mm group 0.20 ± 0.23 mm and 1.42 ± 1.24 Pa respectively (p < 0.001 for both NIT and ESS comparisons across groups). A negative correlation between NIT and baseline ESS was found, the correlation coefficient for all the stented segments was -0.33, p < 0.001. CONCLUSION: In this OCT sub-study of the BIOFLOW II trial, the NIT was statistically different between groups of patients treated with BP-DES and DP-DES. In addition, regions of low ESS were associated with increased NIT in all studied devices.

Non-touch treatment to solve the persistent contamination of culture plates with Aspergillus niger in a clinical microbiology laboratory of Spain.

A pseudo-outbreak of Aspergillus caused by false positive cultures can have a high sanitary impact. We determined the effectiveness (fungal load elimination) of a non-touch disinfection system, vs conventional disinfection methods, to solve steady contamination of culture plates with Aspergillus niger at a clinical microbiology laboratory. Routine cleaning-disinfection (RCD), intensive cleaning-disinfection (ICD), and terminal airborne disinfection (TAD) were employed in stages. Air sampling was carried out before and after each procedure. The effectiveness of TAD on contact surfaces was tested by surface sampling. After RCD, ICD, and TAD, there was a mean decrease of 5.4 (95% CI = 1.8-9.0), 19.2 (95% CI = 12.4-26.0), and 4.4 (95% CI = 2.5-6.3) CFU per tested area, and 46.2%, 21.7%, and 95.5% of contaminated areas became sterile, respectively. There was a mean decrease of 30.6 CFU per tested surface (p < 0.0007) and 50% of tested surfaces became sterile. Global effectiveness of RCD, ICD, and TAD was 68.8% (95% CI = 68.5-69.1), 82.2% (95% CI = 82.1-82.3), and 99.0% (95% CI = 98.8-99.2), respectively. The effectiveness was higher with TAD (4.1 CFU/cm2 less than with ICD; p = 0.0290). No further contamination has occurred since then. When construction and renovation activities are discarded and RCD and ICD practices are insufficient, non-touch disinfection remove residual dust contamination and avoid recurrence.

An anatomically detailed arterial-venous network model. Cerebral and coronary circulation.

In recent years, several works have addressed the problem of modeling blood flow phenomena in veins, as a response to increasing interest in modeling pathological conditions occurring in the venous network and their connection with the rest of the circulatory system. In this context, one-dimensional models have proven to be extremely efficient in delivering predictions in agreement with in-vivo observations. Pursuing the increase of anatomical accuracy and its connection to physiological principles in haemodynamics simulations, the main aim of this work is to describe a novel closed-loop Anatomically-Detailed Arterial-Venous Network (ADAVN) model. An extremely refined description of the arterial network consisting of 2,185 arterial vessels is coupled to a novel venous network featuring high level of anatomical detail in cerebral and coronary vascular territories. The entire venous network comprises 189 venous vessels, 79 of which drain the brain and 14 are coronary veins. Fundamental physiological mechanisms accounting for the interaction of brain blood flow with the cerebro-spinal fluid and of the coronary circulation with the cardiac mechanics are considered. Several issues related to the coupling of arterial and venous vessels at the microcirculation level are discussed in detail. Numerical simulations are compared to patient records published in the literature to show the descriptive capabilities of the model. Furthermore, a local sensitivity analysis is performed, evidencing the high impact of the venous circulation on main cardiovascular variables.

Modeling essential hypertension with a closed-loop mathematical model for the entire human circulation.

Arterial hypertension, defined as an increase in systemic arterial pressure, is a major risk factor for the development of diseases affecting the cardiovascular system. Every year, 9.4 million deaths worldwide are caused by complications arising from hypertension. Despite well-established approaches to diagnosis and treatment, fewer than half of all hypertensive patients have adequately controlled blood pressure. In this scenario, computational models of hypertension can be a practical approach for better quantifying the role played by different components of the cardiovascular system in the determination of this condition. In the present work we adopt a global closed-loop multi-scale mathematical model for the entire human circulation to reproduce a hypertensive scenario. In particular, we modify the model to reproduce alterations in the cardiovascular system that are cause and/or consequence of the hypertensive state. The adaptation does not only affect large systemic arteries and the heart but also the microcirculation, the pulmonary circulation and the venous system. Model outputs for the hypertensive scenario are validated through assessment of computational results against current knowledge on the impact of hypertension on the cardiovascular system.

Quantitative coronary three-dimensional geometry and its association with atherosclerotic disease burden and composition.

BACKGROUND: Isolate features of the coronary anatomy have been associated with the pathophysiology of atherosclerotic disease. Computational methods have been described to allow precise quantification of the complex three-dimensional (3D) coronary geometry. The present study tested whether quantitative parameters that describe the spatial 3D coronary geometry is associated with the extension and composition of the underlying coronary artery disease (CAD). METHODS: Patients with CAD scheduled for percutaneous intervention were investigated with coronary computed tomography angiography (CCTA), and invasive coronary angiography, and virtual histology intravascular ultrasound (IVUS-VH). For all target vessels, 3D centerlines were extracted from CCTA images and processed to quantify 23 geometric indexes, grouped into 3 main categories as follows: (i) length-based; (ii) curvature-based, torsion-based, and curvature/torsion-combined; (iii) vessel path-based. The geometric variables were compared with IVUS-VH parameters assessing the extent and composition of coronary atherosclerosis. RESULTS: A total of 36 coronary patients (99 vessels) comprised the study population. From the 23 geometric indexes, 18 parameters were significantly (p < 0.05) associated with at least 1 IVUS-VH parameter at a univariate analysis. All three main geometric categories provided parameters significantly related with atherosclerosis variables. The 3D geometric indexes were associated with the degree of atherosclerotic extension, as well as with plaque composition. Geometric features remained significantly associated with all IVUS-VH parameters even after multivariate adjustment for clinical characteristics. CONCLUSIONS: Quantitative 3D vessel morphology emerges as a relevant factor associated with atherosclerosis in patients with established CAD.

From fetus to neonate: A review of cardiovascular modeling in early life.

Computational modeling has well-established utility in the study of cardiovascular hemodynamics, with applications in medical research and, increasingly, in clinical settings to improve the diagnosis and treatment of cardiovascular diseases. Most cardiovascular models developed to date have been of the adult circulatory system; however, the perinatal period is unique as cardiovascular physiology undergoes drastic changes from the fetal circulation, during the birth transition, and into neonatal life. There may also be further complications in this period: for example, preterm birth (defined as birth before 37 completed weeks of gestation) carries risks of short-term cardiovascular instability and is associated with increased lifetime cardiovascular risk. Here, we review computational models of the cardiovascular system in early life, their applications to date and potential improvements and enhancements of these models. We propose a roadmap for developing an open-source cardiovascular model that spans the fetal, perinatal, and postnatal periods. This article is categorized under: Cardiovascular Diseases > Computational Models Cardiovascular Diseases > Biomedical Engineering Congenital Diseases > Computational Models.

Effect of High-Intensity Statin Therapy on Atherosclerosis (IBIS-4): Manual Versus Automated Methods of IVUS Analysis.

AIMS: Standard manual analysis of IVUS to study the impact of anti-atherosclerotic therapies on the coronary vessel wall is done by a core laboratory (CL), the ground truth (GT). Automatic segmentation of IVUS with a machine learning (ML) algorithm has the potential to replace manual readings with an unbiased and reproducible method. The aim is to determine if results from a CL can be replicated with ML methods. METHODS: This is a post-hoc, comparative analysis of the IBIS-4 (Integrated Biomarkers and Imaging Study-4) study (NCT00962416). The GT baseline and 13-month follow-up measurements of lumen and vessel area and percent atheroma volume (PAV) after statin induction were repeated by the ML algorithm. RESULTS: The primary endpoint was change in PAV. PAV as measured by GT was 43.95 % at baseline and 43.02 % at follow-up with a change of -0.90 % (p = 0.007) while the ML algorithm measured 43.69 % and 42.41 % for baseline and follow-up, respectively, with a change of -1.28 % (p < 0.001). Along the most diseased 10 mm segments, GT-PAV was 52.31 % at baseline and 49.42 % at follow-up, with a change of -2.94 % (p < 0.001). The same segments measured by the ML algorithm resulted in PAV of 51.55 % at baseline and 47.81 % at follow-up with a change of -3.74 % (p < 0.001). CONCLUSIONS: PAV, the most used endpoint in clinical trials, analyzed by the CL is closely replicated by the ML algorithm. ML automatic segmentation of lumen, vessel and plaque effectively reproduces GT and may be used in future clinical trials as the standard.

Pleiotropic antifibrotic actions of aspirin-triggered resolvin D1 in the lungs.

Introduction: Pulmonary fibrosis is a destructive, progressive disease that dramatically reduces life quality of patients, ultimately leading to death. Therapeutic regimens for pulmonary fibrosis have shown limited benefits, hence justifying the efforts to evaluate the outcome of alternative treatments. Methods: Using a mouse model of bleomycin (BLM)-induced lung fibrosis, in the current work we asked whether treatment with pro-resolution molecules, such as pro-resolving lipid mediators (SPMs) could ameliorate pulmonary fibrosis. To this end, we injected aspirin-triggered resolvin D1 (7S,8R,17R-trihydroxy-4Z,9E,11E,13Z,15E19Z-docosahexaenoic acid; ATRvD1; i.v.) 7 and 10 days after BLM (intratracheal) challenge and samples were two weeks later. Results and discussion: Assessment of outcome in the lung tissues revealed that ATRvD1 partially restored lung architecture, reduced leukocyte infiltration, and inhibited formation of interstitial edema. In addition, lung tissues from BLM-induced mice treated with ATRvD1 displayed reduced levels of TNF-α, MCP-1, IL-1-ß, and TGF-ß. Of further interest, ATRvD1 decreased lung tissue expression of MMP-9, without affecting TIMP-1. Highlighting the beneficial effects of ATRvD1, we found reduced deposition of collagen and fibronectin in the lung tissues. Congruent with the anti-fibrotic effects that ATRvD1 exerted in lung tissues, α-SMA expression was decreased, suggesting that myofibroblast differentiation was inhibited by ATRvD1. Turning to culture systems, we next showed that ATRvD1 impaired TGF-ß-induced fibroblast differentiation into myofibroblast. After showing that ATRvD1 hampered extracellular vesicles (EVs) release in the supernatants from TGF-ß-stimulated cultures of mouse macrophages, we verified that ATRvD1 also inhibited the release of EVs in the bronco-alveolar lavage (BAL) fluid of BLM-induced mice. Motivated by studies showing that BLM-induced lung fibrosis is linked to angiogenesis, we asked whether ATRvD1 could blunt BLM-induced angiogenesis in the hamster cheek pouch model (HCP). Indeed, our intravital microscopy studies confirmed that ATRvD1 abrogates BLM-induced angiogenesis. Collectively, our findings suggest that treatment of pulmonary fibrosis patients with ATRvD1 deserves to be explored as a therapeutic option in the clinical setting.

A modular and reusable model of epithelial transport in the proximal convoluted tubule.

We review a collection of published renal epithelial transport models, from which we build a consistent and reusable mathematical model able to reproduce many observations and predictions from the literature. The flexible modular model we present here can be adapted to specific configurations of epithelial transport, and in this work we focus on transport in the proximal convoluted tubule of the renal nephron. Our mathematical model of the epithelial proximal convoluted tubule describes the cellular and subcellular mechanisms of the transporters, intracellular buffering, solute fluxes, and other processes. We provide free and open access to the Python implementation to ensure our multiscale proximal tubule model is accessible; enabling the reader to explore the model through setting their own simulations, reproducibility tests, and sensitivity analyses.

A computational study of aortic insufficiency in patients supported with continuous flow left ventricular assist devices: Is it time for a paradigm shift in management?

Background: De novo aortic insufficiency (AI) following continuous flow left ventricular assist device (CF-LVAD) implantation is a common complication. Traditional early management utilizes speed augmentation to overcome the regurgitant flow in an attempt to augment net forward flow, but this strategy increases the aortic transvalvular gradient which predisposes the patient to progressive aortic valve pathology and may have deleterious effects on aortic shear stress and right ventricular (RV) function. Materials and methods: We employed a closed-loop lumped-parameter mathematical model of the cardiovascular system including the four cardiac chambers with corresponding valves, pulmonary and systemic circulations, and the LVAD. The model is used to generate boundary conditions which are prescribed in blood flow simulations performed in a three-dimensional (3D) model of the ascending aorta, aortic arch, and thoracic descending aorta. Using the models, impact of various patient management strategies, including speed augmentation and pharmacological treatment on systemic and pulmonary (PA) vasculature, were investigated for four typical phenotypes of LVAD patients with varying degrees of RV to PA coupling and AI severity. Results: The introduction of mild/moderate or severe AI to the coupled RV and pulmonary artery at a speed of 5,500 RPM led to a reduction in net flow from 5.4 L/min (no AI) to 4.5 L/min (mild/moderate) to 2.1 L/min (severe). RV coupling ratio (Ees/Ea) decreased from 1.01 (no AI) to 0.96 (mild/moderate) to 0.76 (severe). Increasing LVAD speed to 6,400 RPM in the severe AI and coupled scenario, led to a 42% increase in net flow and a 16% increase in regurgitant flow (RF) with a nominal decrease of 1.6% in RV myocardial oxygen consumption (MVO2). Blood pressure control with the coupled RV with severe AI at 5,500 RPM led to an 81% increase in net flow with a 15% reduction of RF and an 8% reduction in RV MVO2. With an uncoupled RV, the introduction of mild/moderate or severe AI at a speed of 5,500 RPM led to a reduction in net flow from 5.0 L/min (no AI) to 4.0 L/min (mild/moderate) to 1.8 L/min (severe). Increasing the speed to 6,400 RPM with severe AI and an uncoupled RV increased net flow by 45%, RF by 15% and reduced RV MVO2 by 1.1%. For the uncoupled RV with severe AI, blood pressure control alone led to a 22% increase in net flow, 4.2% reduction in RF, and 3.9% reduction in RV MVO2; pulmonary vasodilation alone led to a 18% increase in net flow, 7% reduction in RF, and 26% reduction in RV MVO2; whereas, combined BP control and pulmonary vasodilation led to a 113% increase in net flow, 20% reduction in RF and 31% reduction in RV MVO2. Compared to speed augmentation, blood pressure control consistently resulted in a reduction in WSS throughout the proximal regions of the arterial system. Conclusion: Speed augmentation to overcome AI in patients supported by CF-LVAD appears to augment flow but also increases RF and WSS in the aorta, and reduces RV MVO2. Aggressive blood pressure control and pulmonary vasodilation, particularly in those patients with an uncoupled RV can improve net flow with more advantageous effects on the RV and AI RF.

Scaling laws and the left main coronary artery bifurcation. A combination of geometric and simulation analyses.

The geometry of coronary arteries is believed to play the role as an atherosclerotic risk factor on its own. The full characterization of the normal coronary network has been reported in the literature. Reports on the integration of geometry and functional data for normal coronary vessels started to proliferate more recently. In this work, we analyze and integrate the geometric data retrieved from angiography images of the left main coronary bifurcation in angiographically normal patients and hemodynamic data generated from blood flow models to analyze the role of allometric laws and the connection between flow distribution and wall shear stress loads on the left anterior descending and left circumflex arteries. This in-silico study contributes to the characterization of normal coronary anatomy and its impact on the hemodynamic shear stresses acting over the vessel wall, shedding light on the impact of geometry-based versus simulation-based hypotheses to define boundary conditions for numerical simulations. We discuss the role of the wall shear stress corresponding to scenarios adopted by the scientific community and the ones proposed in this study. For the simulation-based hypothesis, we propose an iterative strategy to define boundary conditions at the main left coronary bifurcation, such that wall shear stresses are matched between the left descending and left circumflex arteries. From this study, we conclude that a one-fits-all power law exponent of 7/3 results in an good trade-off between computational cost and wall shear stress balance between daughter vessels.

Simultaneous assessment of coronary stenosis relevance with automated computed tomography angiography and intravascular ultrasound analyses and fractional flow reserve.

OBJECTIVES: To assess the diagnostic performance of computed tomography angiography (CTA) and intravascular ultrasound (IVUS) derived minimum lumen areas (MLA) from the same lesions that correspond to an FFR ≤0.80. METHODS AND RESULTS: A total of 24 patients (33 arteries) were collected retrospectively according to the following inclusion criteria: presence of a CTA diagnostic followed by an IVUS and FFR percutaneous coronary procedures. CTA and IVUS lumen contours were automatically performed using previously validated methods.The correlation between CTA and IVUS for the MLA was r = 0.45. In terms of MLA, the mean difference between CTA and IVUS was 0.81 mm2. Of note, a much smaller CTA-derived MLA (2.10 mm2) was found to be related to significant FFR lesions compared to that of the MLA derived from IVUS (3.19 mm2). The area under the curve, accuracy, sensitivity and specificity for this CTA-derived MLA were 0.80, 0.76, 0.50 and 0.87, respectively, while these values for IVUS-derived MLA were 0.87, 0.85, 0.80 and 0.87. CONCLUSIONS: Computed tomography angiography and intravascular ultrasound-derived minimum lumen areas have moderate diagnostic efficiency, albeit slightly better for IVUS, in identifying hemodynamically severe coronary stenoses. The utility of MLA, automatically derived from either CTA or IVUS as an alternative to FFR to guide the decision to revascularize, should be tested clinically.

Fully automated lumen and vessel contour segmentation in intravascular ultrasound datasets.

Segmentation of lumen and vessel contours in intravascular ultrasound (IVUS) pullbacks is an arduous and time-consuming task, which demands adequately trained human resources. In the present study, we propose a machine learning approach to automatically extract lumen and vessel boundaries from IVUS datasets. The proposed approach relies on the concatenation of a deep neural network to deliver a preliminary segmentation, followed by a Gaussian process (GP) regressor to construct the final lumen and vessel contours. A multi-frame convolutional neural network (MFCNN) exploits adjacency information present in longitudinally neighboring IVUS frames, while the GP regression method filters high-dimensional noise, delivering a consistent representation of the contours. Overall, 160 IVUS pullbacks (63 patients) from the IBIS-4 study (Integrated Biomarkers and Imaging Study-4, Trial NCT00962416), were used in the present work. The MFCNN algorithm was trained with 100 IVUS pullbacks (8427 manually segmented frames), was validated with 30 IVUS pullbacks (2583 manually segmented frames) and was blindly tested with 30 IVUS pullbacks (2425 manually segmented frames). Image and contour metrics were used to characterize model performance by comparing ground truth (GT) and machine learning (ML) contours. Median values (interquartile range, IQR) of the Jaccard index for lumen and vessel were 0.913, [0.882,0.935] and 0.940, [0.917,0.957], respectively. Median values (IQR) of the Hausdorff distance for lumen and vessel were 0.196mm, [0.146,0.275]mm and 0.163mm, [0.122,0.234]mm, respectively. Also, the mean value of lumen area predictions, and limits of agreement were -0.19mm2, [1.1,-1.5]mm2, while the mean value and limits of agreement of plaque burden were 0.0022, [0.082,-0.078]. The results obtained with the model developed in this work allow us to conclude that the proposed machine learning approach delivers accurate segmentations in terms of image metrics, contour metrics and clinically relevant variables, enabling its use in clinical routine by mitigating the costs involved in the manual management of IVUS datasets.

Human vs. machine vs. core lab for the assessment of coronary atherosclerosis with lumen and vessel contour segmentation with intravascular ultrasound.

A machine learning (ML) algorithm for automatic segmentation of intravascular ultrasound was previously validated. It has the potential to improve efficiency, accuracy and precision of coronary vessel segmentation compared to manual segmentation by interventional cardiology experts. The aim of this study is to compare the performance of human readers to the machine and against the readings from a Core Laboratory. This is a post-hoc, cross-sectional analysis of the IBIS-4 study. Forty frames were randomly selected and analyzed by 10 readers of varying expertise two separate times, 1 week apart. Their measurements of lumen, vessel, plaque areas, and plaque burden were performed in an offline software. Among humans, the intra-observer variability was not statistically significant. For the total 80 frames, inter-observer variability between human readers, the ML algorithm and Core Laboratory for lumen area, vessel area, plaque area and plaque burden were not statistically different. For lumen area, however, relative differences between the human readers and the Core Lab ranged from 0.26 to 12.61%. For vessel area, they ranged from 1.25 to 9.54%. Efficiency between the ML algorithm and the readers differed notably. Humans spent 47 min on average to complete the analyses, while the ML algorithm took on average less than 1 min. The overall lumen, vessel and plaque means analyzed by humans and the proposed ML algorithm are similar to those of the Core Lab. Machines, however, are more time efficient. It is warranted to consider use of the ML algorithm in clinical practice.

A simple coronary blood flow model to study the collateral flow index.

In this work, we present a novel modeling framework to investigate the effects of collateral circulation into the coronary blood flow physiology. A prototypical model of the coronary tree, integrated with the concept of Collateral Flow Index (CFI), is employed to gain insight about the role of model parameters associated with the collateral circuitry, which results in physically-realizable solutions for specific CFI data. Then, we discuss the mathematical feasibility of pressure-derived CFI, anatomical implications and practical considerations involving the estimation of model parameters in collateral connections. A sensitivity analysis is carried out, and the investigation of the impact of the collateral circulation on FFR values is also addressed.