Multiple resolution residual network for automatic thoracic organs-at-risk segmentation from CT

TL;DR

This paper introduces a multiple resolution residual network (MRRN) that effectively segments thoracic organs-at-risk from CT images, outperforming existing methods especially for challenging structures like the esophagus.

Contribution

The study presents a novel MRRN architecture that combines multi-resolution feature streams with residual connections for improved organ segmentation in thoracic CT scans.

Findings

Outperformed the best method in the AAPM challenge for difficult structures.

Achieved median DSC of 0.97 for lungs, 0.93 for heart, 0.78 for esophagus, and 0.88 for spinal cord.

Demonstrated robustness across a large dataset of thoracic CT scans.

Abstract

We implemented and evaluated a multiple resolution residual network (MRRN) for multiple normal organs-at-risk (OAR) segmentation from computed tomography (CT) images for thoracic radiotherapy treatment (RT) planning. Our approach simultaneously combines feature streams computed at multiple image resolutions and feature levels through residual connections. The feature streams at each level are updated as the images are passed through various feature levels. We trained our approach using 206 thoracic CT scans of lung cancer patients with 35 scans held out for validation to segment the left and right lungs, heart, esophagus, and spinal cord. This approach was tested on 60 CT scans from the open-source AAPM Thoracic Auto-Segmentation Challenge dataset. Performance was measured using the Dice Similarity Coefficient (DSC). Our approach outperformed the best-performing method in the grand challenge for hard-to-segment structures like the esophagus and achieved comparable results for all other structures. Median DSC using our method was 0.97 (interquartile range [IQR]: 0.97-0.98) for the left and right lungs, 0.93 (IQR: 0.93-0.95) for the heart, 0.78 (IQR: 0.76-0.80) for the esophagus, and 0.88 (IQR: 0.86-0.89) for the spinal cord.