Software for stent failure prevention using intravascular OCT images

Summary We will develop an automated software engine for plaque characterization and EEM segmentation using intra- vascular OCT (IVOCT) images, creating a powerful tool in order to efficiently guide stent implantation during PCI. Coronary artery disease (CAD) and its clinical complications are a leading cause of death, disability, and escalating healthcare costs worldwide. Contemporary CAD treatment frequently involves percutaneous coro- nary intervention (PCI) with metal stents in order to resolve obstructive blockages that impede coronary blood flow. While PCI has proven to be effective, subsequent stent failure from restenosis (scar tissue) and throm- bosis (blood clotting) limit the durability of PCI results and are associated with significant morbidity and mortali- ty. Moreover, stent failure has been specifically linked to inadequate stent expansion, incomplete stent cover- age of diseased segments, and untreated dissections at the stent edges.Recently, high-resolution intravascular imaging guidance during PCI has been demonstrated to reduce ad- verse cardiac events by optimizing stent implantation and mitigating structural risks factors for stent failure. However, despite improved patient outcomes, intravascular imaging using ultrasound (IVUS) and optical co- herence tomography (IVOCT) remain severely underutilized in clinical practice. In part, intravascular imaging adoption has been hampered by the need for operators with variable proficiency in image interpretation to per- form manual image analysis on a large volume of data in real-time during the PCI procedure. This creates a scenario where difficulty in image interpretation and an overload of image data (270-500 image frames in a single pull-back) may lead to clinical decision making that relies on incomplete information.We will build upon significant preliminary results and create robust, highly automated methods for identify- ing calcium and lipid deposits in IVOCT image pullbacks, as well as true vessel sizing by automatically deter- mining the location of the EEM. We will: (1) Acquire and label a large, unique dataset of in-vivo IVOCT image volumes. (2) Develop modern machine-learning algorithms for plaque classification and compare against car- diologist readers. (3) Conduct a retrospective validation study to determine how IVOCT with plaque visualiza- tion might affect clinical interventions. (4) Deploy our solution on our cloud platform, LibbyTM, making the soft- ware accessible world-wide, facilitating multinational usage and on-going validation and refinement. We antici- pate that our software will: (1) determine significant lipid and calcium deposits as good as, or better than, ex- pert analysts; (2) incorporate the generated data efficiently into clinical workflow by enhancing pre-PCI imaging to comprehensively map plaque morphology and define appropriate stent landing zones; (3) inform operators on the need for specialized plaque modification techniques such as cutting/scoring balloons or atherectomy; and (4) reliably automate stent sizing by EEM measurements to streamline equipment selection.