Showing papers by "Lucian Mihai Itu published in 2017"

Patient-specific Hemodynamic Computations: Application to Personalized Diagnosis of Cardiovascular Pathologies

Abstract: Hemodynamic computations represent a modern approach for the patientspecific diagnosis of cardiovascular pathologies. These are based on data acquired through medical imaging, and Computational Fluid Dynamics (CFD), which is one of the major topic areas in the field of fluid mechanics, employing numerical methods and algorithms for solving and analyzing applications related to fluid movements. Since the cardiovascular system is a closed-loop system, the simulations performed for particular segment have to take into account the influence of the other cardiovascular components. However, since CFD based computations are computationally expensive, multiscale models have been proposed, which combine detailed threedimensional modeling in the region of interest with oneor zero-dimensional modeling for the remaining components of the cardiovascular system. Finally, parallel processing techniques are typically employed to further reduce the execution time.

Personalized blood flow computations: A hierarchical parameter estimation framework for tuning boundary conditions

Abstract: We propose a hierarchical parameter estimation framework for performing patient-specific hemodynamic computations in arterial models, which use structured tree boundary conditions. A calibration problem is formulated at each stage of the hierarchical framework, which seeks the fixed point solution of a nonlinear system of equations. Common hemodynamic properties, like resistance and compliance, are estimated at the first stage in order to match the objectives given by clinical measurements of pressure and/or flow rate. The second stage estimates the parameters of the structured trees so as to match the values of the hemodynamic properties determined at the first stage. A key feature of the proposed method is that to ensure a large range of variation, two different structured tree parameters are personalized for each hemodynamic property. First, the second stage of the parameter estimation framework is evaluated based on the properties of the outlet boundary conditions in a full body arterial model: the calibration method converges for all structured trees in less than 10 iterations. Next, the proposed framework is successfully evaluated on a patient-specific aortic model with coarctation: only six iterations are required for the computational model to be in close agreement with the clinical measurements used as objectives, and overall, there is a good agreement between the measured and computed quantities. Copyright © 2016 John Wiley & Sons, Ltd.

Semantic segmentation for cancer detection in digital breast tomosynthesis

Abstract: A method, apparatus and non-transitory computer readable medium are for segmenting different types of structures, including cancerous lesions and regular structures like vessels and skin, in a digital breast tomosynthesis (DBT) volume. In an embodiment, the method includes: pre-classification of the DBT volume in dense and fatty tissue and based on the result; localizing a set of structures in the DBT volume by using a multi-stream deep convolutional neural network; and segmenting the localized structures by calculating a probability for belonging to a specific type of structure for each voxel in the DBT volume by using a deep convolutional neural network for providing a three-dimensional probabilistic map.

Towards deep learning based estimation of fracture risk in osteoporosis patients

Abstract: Osteoporosis is a skeletal disorder which leads to bone mass loss and to an increased fracture risk. Recently, physics-based models, employing finite element analysis (FEA), have shown great promise in being able to non-invasively estimate biomechanical quantities of interest in the context of osteoporosis. However, these models have high computational demand, limiting their clinical adoption. In this manuscript, we present a deep learning model based on a convolutional neural network (CNN) for predicting average strain as an alternative to physics-based approaches. The model is trained on a large database of synthetically generated cancellous bone anatomies, where the target values are computed using the physics-based FEA model. The performance of the trained model was assessed by comparing the predictions against physics-based computations on a separate test data set. Correlation between deep learning and physics-based predictions was very good (0.895, p < 0.001), and no systematic bias was found in Bland-Altman analysis. The CNN model also performed better than the previously introduced Support Vector Machine (SVM) model which relied on handcrafted features (correlation 0.847, p < 0.001). Compared to the physics based computation, average execution time was reduced by more than 1000 times, leading to real-time assessment of average strain. Average execution time went down from 32.1 ± 3.0 seconds for the FE model to around 0.03 ± 0.005 seconds for the CNN model on a workstation equipped with 3.0 GHz Intel i7 2-core processor.

Enhanced personalized evaluation of coronary artery disease using an integration of multiple medical imaging techniques

Abstract: A method for providing a personalized evaluation of CAD for a patient includes acquiring one or more non-invasive images depicting a patient's coronary arteries and extracting a first set of features of interest from the one or more non-invasive images. A machine learning model is applied to the first set of features of interest to yield a prediction of one or more coronary measures of interest. One or more invasive images depicting the patient's coronary arteries are acquired and a second set of features of interest are extracted from the one or more invasive images. The first set of features of interest and the second set of features of interest are combined to yield a combined set of features of interest. Then, the machine learning model may be applied to the combined set of features of interest to yield an enhanced prediction of the coronary measures of interest.

Lumped Parameter Whole Body Circulation Modelling

Abstract: In this chapter we introduce methodologies for modeling whole-body cardiovascular dynamics Lumped parameter modeling techniques are employed to model both open-loop and closed-loop dynamics The main constituents of the model are the pulmonary arterial and venous circulation, the systemic arterial and venous circulation, and the four chambers of the heart A fully automated parameter estimation framework is introduced, which is based on two sequential steps: first, a series of parameters are computed directly, and, next, a fully automatic optimization-based calibration method is employed to iteratively estimate the values of the remaining parameters A detailed sensitivity analysis has been performed for identifying the parameters which require calibration Advanced objectives defined based on slopes and interval of times determined from the measured volume and pressure curves are formulated to improve the overall agreement between computed and measured quantities Furthermore, methods for modeling subtle influences, eg from the KG diaphragm, and pathologic heart valves (stenosed, regurgitant) are introduced

Coronary computed tomography clinical decision support system

Abstract: A CT-based clinical decision support system provides coronary decision support. With or without CT-FFR, a machine learnt predictor predicts the clinical decision for the patient based on input from various sources. Using the machine learnt predictor provides more consistent and comprehensive consideration of the available information. The clinical decision support may be provided prior to review of coronary CT data by a radiologist and/or treating physician, providing a starting point or recommendation that may be used by the radiologist and/or treating physician.

Method and System for Personalized Blood Flow Modeling Based on Wearable Sensor Networks

Abstract: A method and system for personalized blood flow modeling based on wearable sensor networks is disclosed. A personalized anatomical model of vessels of a patient is generated based on initial patient data. Continuous cardiovascular measurements of the patient are received from a wearable sensor network on the patient. A computational blood flow model for simulating blood flow in the patient-specific anatomical model of the vessels of the patient is personalized based on the continuous cardiovascular measurements from the wearable sensor network. Blood flow and pressure in the patient-specific anatomical model of the vessels of the patient are simulated using the personalized computational blood flow model. Hemodynamic measures of interest for the patient are computed based on the simulated blood flow and pressure.

Fractional flow reserve decision support system

Abstract: A computed tomography (CT)-based clinical decision support system provides fractional flow reserve (FFR) decision support. The available data, such as the coronary CT data, is used to determine whether to dedicate resources to CT-FFR for a specific patient. A machine-learnt predictor or other model, with access to determinative patient information, is used to assist in a clinical decision regarding CT-FFR. This determination may be made prior to review by a radiologist and/or treating physician to assist decision making.

Patient-Specific Modeling of the Coronary Circulation

Abstract: The instantaneous wave-Free Ratio (IFR) has been recently validated as a rest state pressure-derived index of coronary stenosis severity We demonstrate that IFR and other patient-specific features of rest-state coronary hemodynamics can be quantified via a novel approach based on reduced-order computational modeling of blood flow Blood flow is computed in image-based anatomical reconstructions of the coronary tree from Coronary Angiography (CA) A fully automatic two step parameter estimation framework ensures that the computations match the available patient-specific measurements We evaluate a hybrid decision making strategy c-IFR–invasive FFR against an FFR-only strategy using a dataset comprising 125 lesions (64 patients) Lesions were considered functionally significant if c-IFR 093, while lesions with intermediate c-IFR were classified based on FFR Of the 125 lesions, 43 were hemodynamically significant (FFR ≤ 08) The hybrid c-IFR–FFR strategy resulted in a diagnostic accuracy of 96% when compared to the FFR-only strategy, while requiring invasive FFR assessment in only 34% (43) of the lesions

A Parameter Estimation Framework for Patient-Specific Assessment of Aortic Coarctation

Abstract: In this chapter we introduce a method based on computational fluid dynamics for non-invasively assessing patients with aortic coarctation While in practice the pressure gradient across the coarctation is typically measured invasively with a catheter, the proposed method determines the pressure gradient using a computational modeling approach, which relies on medical imaging data, routine non-invasive clinical measurements and physiological principles The main components of the method are a reduced-order model coupled with a comprehensive pressure-drop formulation, and a parameter estimation method for personalizing the boundary conditions and the vessel wall parameters The parameter estimation method is fully automated, and is based on an iterative tuning procedure to obtain a close match between the computed and the non-invasively determined quantities A key feature is a warm-start to the optimization procedure, with better initial solution for the nonlinear system of equations, to reduce the number of iterations needed for the calibration of the geometrical multiscale models To achieve these goals, the initial solution, computed with a lumped parameter model, is adapted before solving the parameter estimation problem for the geometrical multiscale circulation model: the resistance and the compliance of the circulation model are estimated and compensated This feature is based on research, and is not commercially available Due to regulatory reasons its future availability cannot be guaranteed

Computational Modeling of the Human Cardiovascular System

Abstract: Hemodynamic computations represent a modern approach for the patient-specific diagnosis of cardiovascular pathologies. These are based on data acquired through medical imaging, and Computational Fluid Dynamics (CFD), which is one of the major topic areas in the field of fluid mechanics, employing numerical methods and algorithms for solving and analyzing applications related to fluid movements. Since the cardiovascular system is a closed-loop system, the simulations performed for particular segment have to take into account the influence of the other cardiovascular components. However, since CFD based computations are computationally expensive, multiscale models have been proposed, which combine detailed three-dimensional modeling in the region of interest with one- or zero-dimensional modeling for the remaining components of the cardiovascular system. Finally, parallel processing techniques are typically employed to further reduce the execution time.

GPU-Based High Performance Computing: Employing Massively Parallel Processors for Speeding-Up Compute Intensive Algorithms

Abstract: One-dimensional blood flow models have been used extensively for hemodynamic computations in the human arterial circulation In this chapter we introduce a high performance computing solution based Graphics Processing Units (GPU) Novel GPU only and hybrid CPU-GPU solutions are proposed and evaluated Physiologically sound periodic (structured tree) and non-periodic (windkessel) boundary conditions are considered, in combination with both elastic and viscoelastic arterial wall laws, and different second-order accurate numerical solutions schemes Both the GPU only and the hybrid solutions lead to significantly smaller execution times