ANAGRAPH

by Alvaro Garmendia Navarro

In radiotherapy, individual patient organ shapes and positions are typically extracted from planning CT scans by means of image segmentation. If automated, small errors in segmentation can impair the performance of downstream tasks in the radiotherapy chain. Whenever human validation is not feasible, due to shear scan numbers in research workflows or due to extreeme time restrictions, knowledge based checks become crucial.

In this project our team is exploring knowledge graph representation of the patients anatomical model for individual segmentation predictions of anatomical structures. For head and neck region the motion articulation strongly depends on the skeletal bones and joint connections. Bio-physical motion modelling approaches rely on accurate segmentation quality and anatomical consistency, not always guaranteed by Deep Learning-based prediction models. Graph-based anatomy structures can help to detect mispredictions and their extent - marking predictions suitable towards the modelling requiremets.

Development Team: Alvaro Garmendia Navarro, Richard Häcker, Kristina Giske

Consultants: Oliver Jäkel

Connected projects: CLARITY, PuppetMaster