Deep learning to achieve clinically applicable segmentation of head and neck anatomy for radiotherapy (1809.04430v3)

Abstract: Over half a million individuals are diagnosed with head and neck cancer each year worldwide. Radiotherapy is an important curative treatment for this disease, but it requires manual time consuming delineation of radio-sensitive organs at risk (OARs). This planning process can delay treatment, while also introducing inter-operator variability with resulting downstream radiation dose differences. While auto-segmentation algorithms offer a potentially time-saving solution, the challenges in defining, quantifying and achieving expert performance remain. Adopting a deep learning approach, we demonstrate a 3D U-Net architecture that achieves expert-level performance in delineating 21 distinct head and neck OARs commonly segmented in clinical practice. The model was trained on a dataset of 663 deidentified computed tomography (CT) scans acquired in routine clinical practice and with both segmentations taken from clinical practice and segmentations created by experienced radiographers as part of this research, all in accordance with consensus OAR definitions. We demonstrate the model's clinical applicability by assessing its performance on a test set of 21 CT scans from clinical practice, each with the 21 OARs segmented by two independent experts. We also introduce surface Dice similarity coefficient (surface DSC), a new metric for the comparison of organ delineation, to quantify deviation between OAR surface contours rather than volumes, better reflecting the clinical task of correcting errors in the automated organ segmentations. The model's generalisability is then demonstrated on two distinct open source datasets, reflecting different centres and countries to model training. With appropriate validation studies and regulatory approvals, this system could improve the efficiency, consistency, and safety of radiotherapy pathways.

Deep Learning for Clinically Relevant Segmentation of Head and Neck Anatomy in Radiotherapy

The paper Deep learning to achieve clinically applicable segmentation of head and neck anatomy for radiotherapy presents a comprehensive paper aimed at tackling the challenges of automating organ segmentation in the context of radiotherapy for head and neck cancer. This research utilizes a deep learning architecture, specifically a 3D U-Net model, to achieve expert-level performance in the segmentation of 21 organs at risk (OARs). The aim is to enhance the efficiency, consistency, and safety of radiotherapy treatments by providing reliable automated segmentations that can reduce the variability and time demands associated with manual delineation.

Methodology and Experimental Design

The paper employs a dataset composed of 663 CT scans for the model's training, which includes segmentations derived from clinical practice as well as those manually created by trained radiographers. For evaluation, the model's performance was tested against a separate set of 21 CT scans, each manually segmented by two independent experts, and further validated across two open-source datasets reflecting differing demographics and imaging protocols.

A significant methodological contribution is the introduction of the surface Dice similarity coefficient (surface DSC) metric. Unlike traditional volumetric DSC, the surface DSC evaluates the overlap of OAR surface contours, thus providing a metric more aligned with the clinical task of error correction in automated segmentation. This metric reflects human effort in redrawing segmentation boundaries during clinical workflows, offering a finer-grained performance assessment, particularly pertinent for radiotherapy applications.

Results and Performance Analysis

Quantitatively, the model demonstrates high accuracy in segmenting multiple OARs, achieving performance comparable to experienced radiographers. The surface DSC for the automated segmentations closely matched manual annotations, with notable scores on several critical structures. The model was also subjected to rigorous testing across curated datasets from distinct geographic locations (including the TCIA and PDDCA datasets) to verify its robustness and generalizability.

This research underscores several technical accomplishments:

1. Generalizability: The deep learning model maintained high performance levels across diverse test sets, involving data from different countries and clinical settings, emphasizing robustness in varying imaging conditions.
2. Clinical Relevance: The introduction and application of the surface DSC metric provide a new benchmark for evaluating segmentation tasks in radiotherapy, bridging the metric with actual clinical demands and efforts in post-processing automated outputs.
3. Efficiency: Through automation, the segmentation process for head and neck anatomy can potentially be expedited, leading to more timely radiation therapy planning and potentially reducing the risk of treatment delays.

Implications and Future Directions

The implications of this research are vast, as the automated segmentation model could greatly streamline the radiotherapy planning process by minimizing time-intensive manual segmentation and inter-operator variability. This, in turn, could improve overall treatment quality and patient outcomes. Furthermore, the novel surface DSC metric has the potential to become a standard in evaluating medical image segmentation due to its closer alignment with clinical practice.

Future research areas could explore the integration of other imaging modalities such as MRI, expanding the range of delineable structures, and further refining the surface DSC metric as consensus around its clinical efficacy grows. Additionally, exploring integration with adaptive radiotherapy processes to accommodate anatomical changes over treatment courses is a promising direction.

Ultimately, with continued validation and adherence to rigorous regulatory standards, the deployment of such automated systems in clinical settings could mark a significant step toward enhancing radiotherapy pathways for head and neck cancer patients.