Skip to content

Latest commit

 

History

History
13 lines (7 loc) · 1.55 KB

README.md

File metadata and controls

13 lines (7 loc) · 1.55 KB

Incorporating Geometric Contour Variability into a Dosimetric Radiotherapy Quality Assurance System

This repository contains code to generate data for the ISBI 2024 one-page-abstract poster describing the differences between geometric variability and dosimetric effects. This corresponds to the figure on the left in the image below.

figure-1 drawio

The figure on the right is generated using ASTRA, as shown in the corresponding EMBC '23 paper.

TL;DR: measuring the quality of AI-driven automation in the radiotherapy treatment planning process cannot simply be limited to looking at Dice scores for auto-segmentation, or, DVH curves for deep learning dose calculations, but needs to at least be a combination of both. Current literature is awash with new deep learning network architectures promising a test set Dice of 0.9, which is great! However, do the errors that correspond to the remaining volume of 0.1 impact the dose computation adversely? Uncertainties in a preceding step may propagate into complex failure modes into the next: and a holistic quality assurance system must account for these.

For more, see LinkedIn post here.

The data needed to reproduce this is unfortunately not in the public domain - please reach out to me if you'd like to learn more.