Jessica James, business process analyst and Rebecca Bryan, senior business development manager: in-silico medical solutions lead at the Synopsys Simpleware Product Group, explains recent innovations in using virtual testing and AI to optimise cardiovascular medical devices.
The implanting of cardiovascular devices such as stents and aortic valves typically represents a costly procedure with little room for error. There are also patient-specific challenges for ensuring that devices are suitable for individual anatomies and pathologies, wherein a ‘one-size-fits-all’ approach has limitations. One recent solution being adopted by medical device researchers is to use 3D image processing to accurately represent complex anatomies and device performance.
In silico testing and in silico clinical trials aid in evaluating medical devices and better understanding physiological interactions as a lower-cost, efficient complement to benchtop, animal, or clinical trials when planning products and regulatory submissions. Computational modeling and simulation allow researchers, engineers, and clinicians to investigate specific features of, for example, the heart and cardiovascular system, and run multiple design iterations to learn more about optimal placement and predicted performance under different conditions.
Looking in particular at work carried out at MIT and through a consortium made up of Synopsys, Johns Hopkins University, Thornton Tomasetti, and Duke University, how are organisations tackling specific types of stent challenges and potentially improving patient outcomes? In addition, how are AI-enabled technologies helping with these workflows?
Tackling thrombosis formation in malapposed coronary stents
Researchers at MIT working with Harvard Medical School are exploring new methods to investigate stent thrombosis, or clotting. With over 500,000 coronary stents implanted in the USA each year as a response to coronary artery disease, stent thrombosis is a complication where thrombotic material is created that can result in post-intervention occlusion. Benchtop flow testing has traditionally been used to study how clots form under hemodynamic conditions, but this is limited in its ability to replicate in-vivo conditions, while clinical testing is not feasible due to the unanticipated nature of thrombosis.
An in vitro flow loop setup was used by MIT to simulate blood flow conditions similar to those in a human coronary artery. Stents were deployed within the flow loops under a range of under-expansion conditions. The next stage in the project involved creating in vitro flow loops with metal stents and blood to model thrombosis formation for different types of malapposition in a physical model. Stent malapposition describes a gap between the surface of a stent strut and the arterial wall that is larger than the actual strut thickness.
Samples were scanned using micro-computed tomography (micro-CT) to create DICOM files for 3D image processing in Synopsys Simpleware software. Precomputed thresholding levels from prior experimental tests were used to segment stent struts from the fluid volume and clot formation, while smoothing filters were utilised to obtain continuous structures as part of a 3D visualisation workflow. The Simpleware API was later used to extract pixel values for each mask for each micro-CT slice, which were then plotted as an indication of clot formation. Finally, a custom MATLAB program was employed to further extract strut position and calculate wall distances from mask pixel values on each slice.
Metrics of clot and stent-wall malapposition as a function of stent length were obtained as 1-D signals and studied through frequency domain analysis and magnitude squared coherence. Stent-clot patterns were observed, and the location of malapposition identified, including the correlation of strut and clot location. By using micro-CT imaging and signal analysis, additional insights could be gained to complement benchtop quantification of thrombosis, including the position of stents and reactivity of blood. The 3D computational model also enables flow simulations to better understand local and regional organisation of the stent, flow, and anatomy. Looking to the future, this method promises to generate valuable data on how to define different thrombotic mechanisms and reduces risk when deciding how to place stents for patients.
Testing in silico workflows from 4D CT to FEA
More recent work in cardiovascular device simulation involves the use of AI-enabled technologies to speed up complex 3D and 4D image processing when dealing with cardiovascular devices. In particular, Synopsys, Thornton Tomasetti, Johns Hopkins University, and Duke University have collaborated on a project whereby an in silico workflow was developed using automatic segmentation, mesh morphing, and Finite Element Analysis (FEA) to model aortic valve frame deployment.
For this particular study, a workflow was developed for an in silico study of patient-specific heart valve frame deployments. Anonymised 4D CT patient data was provided by Duke University consisting of 3D volumetric images with 10 time frames per subject, representing a full cardiac cycle. Synopsys Simpleware AS Cardio software was used to import the image data and carry out automated segmentation and landmarking through AI-enabled Machine Learning algorithms, generating a set of meshed models for further analysis.
Compared to the hours typically spent on these image processing tasks, the Simpleware AI algorithm creates usable models in just three minutes pre frame, and allows researchers to almost completely automate a time-consuming process. Johns Hopkins University subsequently used the segmentations and landmarks to morph the models from time frame to time frame to model the cardiac motion, creating consistency for analysing and visualising images within a single coordinate system.
Finally, the mesh data created by Simpleware software from the 4D CT data was employed by Thornton Tomasetti to simulate heart frame deployment in Abaqus software for five patient datasets. It was possible to model a wide range of different properties and patient scenarios to better understand cardiac and vascular motion for the implanted aortic valve frames. Insights were therefore gained into patient-specific aortic transcatheter heart valve analysis and performance characteristics of the aortic valve frames.
Conclusion
Both cases demonstrate recent and ongoing advances in using 3D imaging, AI, and computational techniques to enhance traditional benchtop testing and widen the options available for cardiovascular device manufacturers. More broadly, in silico studies and trials complement traditional methods of testing and supporting regulatory submissions, allowing new devices to potentially reach the market faster with more understanding of their expected performance, suitability and safety. Clinicians can use the results gained by these studies to support clinical decision-making and lead to safer and more optimised diagnostic and surgical solutions for patients, particularly in cases where small variations can make a significant difference to outcomes.
