Abstract:Accurate delineation of clinical target volume (CTV) is one of the most crucial aspects of treatment planning in radiation therapy. However, the quality of this process depends on the expertise level of the individual observer. This study aimed to investigate whether deep-learning based automatic delineation could provide greater accuracy, inter-observer consistency and efficiency of CTV contour compared with manual delineation for pN2 NSCLC receiving post-operation radiation therapy (PORT). A very deep dilated residual network was used to achieve the effective automatic delineation of CTV. The dilated module with different atrous rates is able to extract multi-scale features from CT, leading to a greater robustness of the model. Eleven junior radiation oncologists (work experience ≤ 5 years) delineated CTV on 19 patients using two contour methods: (1) a manual contour from scratch (MC) and (2) a user adjustment of the deep-learning based auto-delineation contour (UADC). Three senior radiation oncologists (work experience > 10 years) also contoured the CTV on these patients and the majority voting was used to generate the ground truth (GT). The accuracy of junior’s delineation was evaluated in terms of mean distance to agreement (MDA) and Dice similarity coefficient (DSC) with the GT as reference. MDA is a distance parameter while DSC is a volume overlap index. Therefore, smaller MDA and higher DSC indicate a greater contour accuracy. The coefficient of variation (COV) was rendered to measure the inter-observer variability. COV is defined as the standard deviation (SD) divided by the mean CTV volume of all observers and a larger COV implies a greater variability. For each of the CTV contouring tasks, the time consumption was recorded. A total of 418 unique CTV sets were generated. The UADC presented a significantly smaller MDA (mm) per individual CTV set compared with MC (mean ± SD: 2.79 ± 0.91 vs. 3.07 ± 0.98, P < 0.001). Similarly, the DSC of UADC was also significantly greater than that of MC (mean ± SD: 0.75 ± 0.06 vs. 0.72 ± 0.07, P < 0.001). For inter-observer variability, UADC introduced a remarkably decreased COV in comparison with MC (mean ± SD: 0.129 ± 0.040 vs. 0.183 ± 0.043, P < 0.001). The median contouring time for MC and UADC was 14.99 min (P25, P75: 9.80, 20.87) and 10.42 min (P25, P75: 7.20, 17.03) respectively, resulting in a 31% reduction (absolute: 4.57 min) of time consumption (P < 0.001). Compared with the manual contour, user adjustment of deep-learning based auto-delineation of PORT-CTV for pN2 NSCLC is a promising strategy to offer superior accuracy and consistency within a shortened time span, leading to higher quality and efficiency of CTV contouring.

Improving the Accuracy and Consistency in Clinical Target Volume Delineation for Rectal Carcinoma by an Education Program

Volumetric and Dosimetric Variations in Radiation Treatment Planning Using CT Images Based on Different Reconstructed Field-of-view

Comparison of Computed Tomography- and Magnetic Resonance Imaging-Based Target Delineation for Cervical Cancer Brachytherapy

Postoperative Radiotherapy For Glioma: Improved Delineation Of The Clinical Target Volume Using The Geodesic Distance Calculation

Interobserver variability in target volume delineation in definitive radiotherapy for thoracic esophageal cancer: a multi-center study from China

Clinical evaluation of the convolutional neural network‑based automatic delineation tool in determining the clinical target volume and organs at risk in rectal cancer radiotherapy

Clinical Application and Improvement of a CNN-based Autosegmentation Model for Clinical Target Volumes in Cervical Cancer Radiotherapy

Inter-observer variation of target volume delineation for CT-guided cervical cancer brachytherapy

A Multicentral Quality Assurance Study of Target Volume Delineation for A Prospective, Random, Multicenter Study: 3JECOG P-01

Inter-observer and Intra-observer Variability in Gross Tumor Volume Delineation of Primary Esophageal Carcinomas Based on Different Combinations of Diagnostic Multimodal Images.

Improved Prostate Delineation in Prostate Hdr Brachytherapy with Trus-Ct Deformable Registration Technology: A Pilot Study with Mri Validation

Evaluating the relationship between contouring variability and modelled treatment outcome for prostate bed radiotherapy

Exploring the Interobserver Agreement in Computer-Aided Radiologic Tumor Measurement and Evaluation of Tumor Response

Deep-learning Based Automatic Delineation Improves CTV Contouring Quality and Efficiency for Pathological N2 (pn2) Non-small Cell Lung Cancer (NSCLC) Receiving Post-operation Radiation Therapy

Clinical evaluation on automatic segmentation results of convolutional neural networks in rectal cancer radiotherapy

Automatic delineation of the clinical target volume and organs at risk by deep learning for rectal cancer postoperative radiotherapy

Investigating rectal toxicity associated dosimetric features with deformable accumulated rectal surface dose maps for cervical cancer radiotherapy

Interobserver Variations in Target Delineation in Intensity-Modulated Radiation Therapy for Nasopharyngeal Carcinoma and Its Impact on Target Dose Coverage

Variations of Clinical Target Volume Delineation for Primary Site of Nasopharyngeal Cancer Among Five Centers in China

Prospective International Pilot Study Evaluating the Efficacy of a Self-Guided Contouring Teaching Module with Integrated Feedback for Transitioning from 2D to 3D Treatment Planning

Evaluation of Deep Learning‐based Auto‐segmentation Algorithms for Delineating Clinical Target Volume and Organs at Risk Involving Data for 125 Cervical Cancer Patients