Art care: A multi-modality coronary 3D reconstruction and hemodynamic status assessment software

1. Introduction

The continuous increase in the mortality rate due to cardiovascular disease (CVD) and more particularly due to coronary artery disease (CAD) has constituted the development of methods that allow the anatomic and functional assessment of coronary vessels of utmost importance. Several imaging modalities have been developed and are used in everyday clinical practice, such as traditional invasive coronary angiography (ICA), coronary CT angiography (CTA), cardiac magnetic resonance imaging (MRI), transthoracic echocardiography, optical coherence tomography (OCT) and IntraVascular UltraSound (IVUS). Moreover, several dedicated 3D-reconstruction and coronary functional assessment commercial software packages have been developed during the past few years which utilize either one (ICA [1] or CCTA [2]) or two imaging modalities (IVUS-ICA [3] or OCT-ICA [4]). However, with the exception of HeartFlow ® [2], none of the aforementioned systems provides the ability to perform the necessary blood flow calculations in order to have a full perspective of the true hemodynamic status of the examined coronary vessel.

We have developed a software platform that provides the clinician with the ability to reconstruct in 3D the desired vessel using three different imaging modality options: fusion of IVUS and ICA, fusion of OCT and ICA or just ICA (3D-QCA). The proposed system offers numerous 3D visualization options and has the ability to use any given IVUS, OCT or ICA sequence. The key point, however, is that using a dedicated finite element module, the virtual functional assessment index (vFAI) [5] for the reconstructed model can be calculated, thus offering both anatomic characteristics, as well as functional assessment of the diseased vessel. The system features have been adequately validated, presenting promising results regarding the automatic lumen border detection from IVUS images and from OCT images, as well as the 3D reconstruction from the 3D-QCA module. Moreover, the calculated vFAI deriving from the 3D reconstructed models using the proposed method presented high correlation when compared to the respective calculated vFAI values deriving from the 3D models reconstructed using the CAAS QCA 3D ® [1] commercial package.

2. Materials and methods

The system incorporates three 3D reconstruction subsystems, regarding the desired imaging modalities (IVUS/Angiography, OCT/Angiography, 3D-QCA). The created 3D model is visualized and then it can be subjected to blood flow simulations, resulting to the calculation of the vFAI. Regarding the IVUS-Angiography and the OCT-Angiography use cases, the user can manually select and extract the end-diastolic (R-peak) frames to perform their segmentation. The luminal borders are manually or automatically annotated in the IVUS or the OCT frames. The user can also trace the outer borders of the External Elastic Membrane (EEM) manually or automatically, in order to create the respective outer wall 3D model. The centerline extraction module is common for all the reconstruction modules. Regarding the 3D-QCA use case, the system utilizes the luminal borders from the centerline extraction module and creates the respective contours for the final 3D model. The 3D models that are created can be then subjected to computational blood flow simulations by an integrated finite element open-source program (PAK-F) [6]. A penalty formulation is also used in the solver [7]. The software automatically creates a 3D 8-node finite element mesh of around 150000 elements (depending on the length of the arterial segment of interest), after the creation of the final 3D model. The user initially sets the parameters for the material properties blood (i.e. density and viscosity). The predefined values are 1050 kg/m 3 for the density and 0.0035 Pa ⋅ s for the dynamic viscosity, respectively. Subsequently, the required blood flow simulations are performed, in order to calculate the virtual Functional Assessment Index (vFAI). These simulations are carried out under the assumptions that the flow is laminar and incompressible, that blood behaves as a Newtonian fluid and that the arterial walls are rigid. All of the system modules are integrated into a user-oriented graphical interface. The graphical interface of the system is depicted in Fig. 2. The system was developed using C++ [8], combined with Microsoft ® .NET Framework 3.5. Furthermore, several other 3 rd party libraries were used for the system implementation. DCMTK [9] was used to extract images and metadata from DICOM files and OpenCV [10] was used for the processing of 2D images. The 3D visualization module was developed using the OpenGL and the VTK libraries, respectively. The system requires a minimum setup composed of Windows XP ® , C++ and .NET Redistributables (2005/2008), an Intel Core Duo processor and 4 GB of RAM. The overall system architecture is depicted in Fig. 1.

OPEN IN VIEWER

OPEN IN VIEWER

In order to validate the proposed angiography reconstruction method (3D-QCA), we compared the implemented algorithm to an already validated commercial 3D-QCA reconstruction software (CAAS QCA 3D ® ) using as validation metrics the volume of the 3D reconstructed models, the length and the minimum lumen diameter of the examined arterial segments and the calculated vFAI on eleven coronary arterial segments.

Strong correlation was found between the two methods for the used validation metrics. In particular, the Pearson’s correlation coefficient (R) for the calculated volumes, vFAI, length and minimum lumen diameter was 0.99, 0.99, 0.99 and 0.88, respectively. The respective mean differences of the aforementioned metrics were 3.53 mm 3 (SD = 8.07 mm 3 ), 0.0146 (SD = 0.021 mm), - 1.99 mm (SD = 2.94 mm) and - 0.034 mm (SD = 0.163 mm), respectively (Fig. 3).

OPEN IN VIEWER

Good agreement was also found for the two methods by the Bland-Altman method of analysis. In fact, the corresponding limits of agreement for the calculated volumes, vFAI, length and minimum lumen diameter were from - 12.3 to 19.3, from - 0.027 to 0.056, from - 7.75 to 3.77 and from - 0.35 to 0.29, respectively.

We also validated the automatic border detection for the IVUS and the OCT modules by comparing them to the respective manually annotated borders made by experienced clinicians. Five pullbacks acquired from five different patients were used to validate the proposed IVUS border detection step and six pullbacks were acquired from six different patients and used to validate the proposed OCT border detection method. A medical expert selected randomly 632 images from the IVUS dataset and 556 images from the OCT dataset and manually annotated the lumen borders. Images having noise artifacts were not included in the validation procedure. The validation results are depicted in Fig. 3.

We have presented a newly developed software that combines the ability to reconstruct in 3D coronary arterial segments using three different sets of imaging modalities and to calculate the vFAI on the segments of interest. The novelty of the proposed system is that it can assess the hemodynamic status of a coronary arterial segment in a matter of minutes, utilizing the most well-known coronary imaging modalities, without the use of a dedicated pressure wire. This leads to the reduction of the total cost of the diagnostic examinations and gives the opportunity to the clinician for a quick and valid decision on the forthcoming type of treatment for the patient. When compared to other publically available software suites, the proposed system offers several advantages since it is the only one that can utilize all three coronary imaging modalities (IVUS-OCT-ICA) and provides the ability to perform blood flow simulations and assess the hemodynamic status of the vessels of interest. The total time required for a full 3D reconstruction and the subsequent vFAI calculation varies according to the available modality (2 minutes for the 3D-QCA module, 10 minutes for the IVUS-ICA and OCT-ICA modules).

Moreover, the automatic OCT and IVUS border detection modules decrease the total reconstruction time even more, since their results are very promising. One other aspect worth mentioning is the fact that the IVUS border detection algorithm performs equally using either of the two IVUS catheters (Volcano or Boston Scientific). Regarding the calculation of the vFAI, strong correlation was observed when compared to the respective 3D models deriving from CAAS QCA 3D ® that were previously validated using their actual FFR measurements, thus proving its efficacy.

We should, however, state that the inability of the implemented algorithms to include bifurcations in the final 3D models is the main limitation of the proposed system. Moreover, we are currently developing a new, larger dataset to further validate the 3D-QCA module with a wider variety of cases regarding the severity of the lesions.