Quantitative framework to predict CTEPH surgical outcome from imaging

点击翻译按钮获取中文摘要

Project Summary The proposal “Quantitative framework to predict CTEPH surgical outcome from imaging” has a long term objective of improving matching of Chronic Thromboembolic Pulmonary Hypertension (CTEPH) patients to their optimum therapy. Currently, advancement of this goal is limited by the lack of quantitative tools and metrics available to physicians to standardize evaluation of patient disease seen on imaging. In this proposal, we aim to tackle two different aspects of this problem. First, we aim to develop metrics to comprehensively quantify disease from imaging in a manner that informs disease severity. In this first aim, we are using dual- energy CT images to capture, from a single study, both the amount and location of vascular obstruction, perfusion deficit, and their relationship to one another. These metrics will be robustly designed to incorporate all levels of the vasculature (proximal to distal), to capture a range of occlusion severities, and to use location weightings based on surgical treatment accessibility. The utility of the metrics will be in their ability to inform both pre and post operative invasive hemodynamics. Our second aim of the proposal is to utilize CT pulmonary angiograms to predict the surgical accessibility of patient disease. We will train convolutional neural networks to predict the vascular location (and therefore surgical accessibility) of CTEPH using the UCSD surgical disease level classification. Neural networks will greatly aid in systematic prediction of disease location, since they can analyze images without data loss, and can also incorporate both clinical and imaging data. Because UCSD performs the highest volume of pulmonary thromboendarterectomy surgeries (a surgery to remove the CTEPH vascular obstructions) in the world, we are the only institution that has the required number of pre- operative images and gold standard (surgically confirmed) assessed surgical disease level classifications to train and evaluate a neural network approach. In future work, these tools can be combined to rapidly, systematically, and quantitatively evaluate CTEPH patients. With these metrics that standardize evaluation, we will be able to quantify factors that contribute to CTEPH phenotypes and determine which of these imaging phenotypes are most responsive to surgery.