Automatic identification of segmentation errors for radiotherapy using geometric learning

TL;DR

This paper introduces a novel AI-based tool combining CNN and GNN to automatically detect segmentation errors in CT scans for radiotherapy, aiming to reduce manual editing time.

Contribution

It presents a new architecture leveraging appearance and shape features, trained with self-supervised learning on synthetic data, for error detection without ground truth.

Findings

Achieved 85-89.7% precision in error prediction

Attained 66.5-68.6% recall for internal and external errors

Demonstrated potential to streamline clinical workflow

Abstract

Automatic segmentation of organs-at-risk (OARs) in CT scans using convolutional neural networks (CNNs) is being introduced into the radiotherapy workflow. However, these segmentations still require manual editing and approval by clinicians prior to clinical use, which can be time consuming. The aim of this work was to develop a tool to automatically identify errors in 3D OAR segmentations without a ground truth. Our tool uses a novel architecture combining a CNN and graph neural network (GNN) to leverage the segmentation's appearance and shape. The proposed model is trained using self-supervised learning using a synthetically-generated dataset of segmentations of the parotid and with realistic contouring errors. The effectiveness of our model is assessed with ablation tests, evaluating the efficacy of different portions of the architecture as well as the use of transfer learning from an unsupervised pretext task. Our best performing model predicted errors on the parotid gland with a precision of 85.0% & 89.7% for internal and external errors respectively, and recall of 66.5% & 68.6%. This offline QA tool could be used in the clinical pathway, potentially decreasing the time clinicians spend correcting contours by detecting regions which require their attention. All our code is publicly available at https://github.com/rrr-uom-projects/contour_auto_QATool.