WP11 – National Quality Assurance program for radiotherapy delineation

The study is a national multi-centre from the four neuro-oncology centres in Denmark in Odense, Copenhagen, Aarhus, and Aalborg and the Danish Centre of Particle Therapy.

Rationale

High precision RT techniques depend on steep dose gradients and delineation errors may thus lead to inadequate dose coverage reducing the chance of tumour control or cause unnecessary toxicity. Retrospective analyses of EORTC intergroup trials on low grade glioma and anaplastic glioma revealed that the delineation of target volumes and organs at risk are common causes of protocol deviations in clinical trials [35,36]

Objective

The overall aim is to prospectively ensure a national high quality of normal tissue and target definition through use of advanced automatic semantic segmentation methods and continuous centralized quality assurance of these structures in the brain.

Materials and methods

The objective will be achieved by

1) collecting all future delineation and relevant imaging of brain cancer patients into a national database (dcmCollab)

2) train a Convolutional Neural Network to semantic segment normal tissue and target

3) validate this CNN pipeline with expert delineations tested against a national interobserver delineation study from the DCCC RT workgroup (Figure 10)

4) Once validated all future national delineation will be compared to the Deep Learning semantic segmentation as a reference and potentially identify subpar contours in the national context.

Endpoints

1) Establishing a DNOG validated Deep Learning semantic segmentation pipeline.

2) Automatic identification of subpar contours.

Expected Impact. Increased quality of radiotherapy