Discrete Pseudohealthy Synthesis: Aortic Root Shape Typification and Type Classification with Pathological Prior
Keywords: clustering, representation learning, personalized prosthetics, valve-sparing aortic root reconstruction
TL;DR: We present a method to automatically identify a set of healthy shape types and estimate the optimal one given a pathological image, aiming on personalized prostheses shaping.
Abstract: In personalized prosthesis shaping, the desired shape remains typically unknown and has to be estimated based on the individual pathological shape. This estimation is also called pseudo healthy synthesis. One example application is the personalization of aortic root prostheses during valve-sparing aortic root surgery. Even though several methods for pseudohealthy synthesis were proposed during the last years, it might not always be necessary to taylor a completely individual and unique prosthesis for each and every patient as this introduces high costs and regulatory issues. Another option is to identify a set of prosthesis types that represents all natural healthy shapes in an adequate way. Then, the pseudohealthy synthesis problem becomes a classification problem, aiming on predicting the optimal prosthesis out of the set of candidates given a pathological shape. In this work, we present a fully automized workflow of unsupervised shape typification and type classification based on pathological data for the example of personalizing aortic root prostheses shapes. We provide a proof-of-concept study on an ex-vivo porcine data set, including a thorough evaluation of the model's hyperparameters and the number of identified shape types. Our study lies the groundwork for a new branch of personalized prosthesis shaping with a high potential of translation to clinical application: Discrete Pseudohealthy Synthesis.
Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.
Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.
Paper Type: validation/application paper
Source Latex: zip
Primary Subject Area: Unsupervised Learning and Representation Learning
Secondary Subject Area: Application: Other