Dataset Information

Feasibility of Continual Deep Learning-Based Segmentation for Personalized Adaptive Radiation Therapy in Head and Neck Area.

ABSTRACT: This study investigated the feasibility of deep learning-based segmentation (DLS) and continual training for adaptive radiotherapy (RT) of head and neck (H&N) cancer. One-hundred patients treated with definitive RT were included. Based on 23 organs-at-risk (OARs) manually segmented in initial planning computed tomography (CT), modified FC-DenseNet was trained for DLS: (i) using data obtained from 60 patients, with 20 matched patients in the test set (DLSm); (ii) using data obtained from 60 identical patients with 20 unmatched patients in the test set (DLSu). Manually contoured OARs in adaptive planning CT for independent 20 patients were provided as test sets. Deformable image registration (DIR) was also performed. All 23 OARs were compared using quantitative measurements, and nine OARs were also evaluated via subjective assessment from 26 observers using the Turing test. DLSm achieved better performance than both DLSu and DIR (mean Dice similarity coefficient; 0.83 vs. 0.80 vs. 0.70), mainly for glandular structures, whose volume significantly reduced during RT. Based on subjective measurements, DLS is often perceived as a human (49.2%). Furthermore, DLSm is preferred over DLSu (67.2%) and DIR (96.7%), with a similar rate of required revision to that of manual segmentation (28.0% vs. 29.7%). In conclusion, DLS was effective and preferred over DIR. Additionally, continual DLS training is required for an effective optimization and robustness in personalized adaptive RT.

SUBMITTER: Kim N

PROVIDER: S-EPMC7915955 | biostudies-literature | 2021 Feb

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Feasibility of Continual Deep Learning-Based Segmentation for Personalized Adaptive Radiation Therapy in Head and Neck Area.

Kim Nalee N Chun Jaehee J Chang Jee Suk JS Lee Chang Geol CG Keum Ki Chang KC Kim Jin Sung JS

Cancers 20210209 4

This study investigated the feasibility of deep learning-based segmentation (DLS) and continual training for adaptive radiotherapy (RT) of head and neck (H&N) cancer. One-hundred patients treated with definitive RT were included. Based on 23 organs-at-risk (OARs) manually segmented in initial planning computed tomography (CT), modified FC-DenseNet was trained for DLS: (i) using data obtained from 60 patients, with 20 matched patients in the test set (DLSm); (ii) using data obtained from 60 ident ...[more]

PMID: 33572310

Similar Datasets

Project description:BackgroundThe aim of the present study is to examine the impact of kV-CBCT-based online adaptive radiation therapy (ART) on dosimetric parameters in comparison to image-guided-radiotherapy (IGRT) in consecutive patients with tumors in the head and neck region from a prospective registry.MethodsThe study comprises all consecutive patients with tumors in the head and neck area who were treated with kV-CBCT-based online ART or IGRT-modus at the linear-accelerator ETHOS™. As a measure of effectiveness, the equivalent-uniform-dose was calculated for the CTV (EUDCTV) and organs-at-risk (EUDOAR) and normalized to the prescribed dose. As an important determinant for the need of ART the interfractional shifts of anatomic landmarks related to the tongue were analyzed and compared to the intrafractional shifts. The latter determine the performance of the adapted dose distribution on the verification CBCT2 postadaptation.ResultsAltogether 59 consecutive patients with tumors in the head-and-neck-area were treated from 01.12.2021 to 31.01.2023. Ten of all 59 patients (10/59; 16.9%) received at least one phase within a treatment course with ART. Of 46 fractions in the adaptive mode, irradiation was conducted in 65.2% of fractions with the adaptive-plan, the scheduled-plan in the remaining. The dispersion of the distributions of EUDCTV-values from the 46 dose fractions differed significantly between the scheduled and adaptive plans (Ansari-Bradley-Test, p = 0.0158). Thus, the 2.5th percentile of the EUDCTV-values by the adaptive plans amounted 97.1% (95% CI 96.6-99.5%) and by the scheduled plans 78.1% (95% CI 61.8-88.7%). While the EUDCTV for the accumulated dose distributions stayed above 95% at PTV-margins of ≥ 3 mm for all 8 analyzed treatment phases the scheduled plans did for margins ≥ 5 mm. The intrafractional anatomic shifts of all 8 measured anatomic landmarks were smaller than the interfractional with overall median values of 8.5 mm and 5.5 mm (p < 0.0001 for five and p < 0.05 for all parameters, pairwise comparisons, signed-rank-test). The EUDOAR-values for the larynx and the parotid gland were significantly lower for the adaptive compared with the scheduled plans (Wilcoxon-test, p < 0.001).ConclusionsThe mobile tongue and tongue base showed considerable interfractional variations. While PTV-margins of 5 mm were sufficient for IGRT, ART showed the potential of decreasing PTV-margins and spare dose to the organs-at-risk.

Project description:PurposeTo identify which patient-reported outcomes (PROs) may be most improved through adaptive radiation therapy (ART) with the goal of reducing toxicity incidence among head and neck cancer patients.MethodsOne hundred fifty-five head and neck cancer patients receiving radical VMAT (chemo)radiotherapy (66-70 Gy in 30-35 fractions) completed the MD Anderson Symptom Inventory, MD Anderson Dysphagia Inventory (MDADI), and Xerostomia Questionnaire while attending routine follow-up clinics between June-October 2019. Hierarchical clustering characterized symptom endorsement. Conventional statistical approaches indicated associations between dose and commonly reported symptoms. These associations, and the potential benefit of interfractional dose corrections, were further explored via logistic regression.ResultsRadiotherapy-related symptoms were commonly reported (dry mouth, difficulty swallowing/chewing). Clustering identified three patient subgroups reporting: none/mild symptoms for most items (60.6% of patients); moderate/severe symptoms affecting some aspects of general well-being (32.9%); and moderate/severe symptom reporting for most items (6.5%). Clusters of PRO items broadly consisted of acute toxicities, general well-being, and head and neck-specific symptoms (xerostomia, dysphagia). Dose-PRO relationships were strongest between delivered pharyngeal constrictor Dmean and patient-reported dysphagia, with MDADI composite scores (mean ± SD) of 25.7 ± 18.9 for patients with Dmean <50 Gy vs. 32.4 ± 17.1 with Dmean ≥50 Gy. Based on logistic regression models, during-treatment dose corrections back to planned values may confer ≥5% decrease in the absolute risk of self-reported physical dysphagia symptoms ≥1 year post-treatment in 1.2% of patients, with a ≥5% decrease in relative risk in 23.3% of patients.ConclusionsPatient-reported dysphagia symptoms are strongly associated with delivered dose to the pharyngeal constrictor. Dysphagia-focused ART may provide the greatest toxicity benefit to head and neck cancer patients, and represent a potential new direction for ART, given that the existing ART literature has focused almost exclusively on xerostomia reduction.

Project description:BackgroundOver half a million individuals are diagnosed with head and neck cancer each year globally. Radiotherapy is an important curative treatment for this disease, but it requires manual time to delineate radiosensitive organs at risk. This planning process can delay treatment while also introducing interoperator variability, resulting in downstream radiation dose differences. Although auto-segmentation algorithms offer a potentially time-saving solution, the challenges in defining, quantifying, and achieving expert performance remain.ObjectiveAdopting a deep learning approach, we aim to demonstrate a 3D U-Net architecture that achieves expert-level performance in delineating 21 distinct head and neck organs at risk commonly segmented in clinical practice.MethodsThe model was trained on a data set of 663 deidentified computed tomography 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 organ at risk definitions.ResultsWe demonstrated the model's clinical applicability by assessing its performance on a test set of 21 computed tomography scans from clinical practice, each with 21 organs at risk segmented by 2 independent experts. We also introduced surface Dice similarity coefficient, a new metric for the comparison of organ delineation, to quantify the deviation between organ at risk surface contours rather than volumes, better reflecting the clinical task of correcting errors in automated organ segmentations. The model's generalizability was then demonstrated on 2 distinct open-source data sets, reflecting different centers and countries to model training.ConclusionsDeep learning is an effective and clinically applicable technique for the segmentation of the head and neck anatomy for radiotherapy. With appropriate validation studies and regulatory approvals, this system could improve the efficiency, consistency, and safety of radiotherapy pathways.

Project description:Purpose Fast and accurate auto-segmentation on daily images is essential for magnetic resonance imaging (MRI)–guided adaptive radiation therapy (ART). However, the state-of-the-art auto-segmentation based on deep learning still has limited success, particularly for complex structures in the abdomen. This study aimed to develop an automatic contour refinement (ACR) process to quickly correct for unacceptable auto-segmented contours. Methods and Materials An improved level set–based active contour model (ACM) was implemented for the ACR process and was tested on the deep learning–based auto-segmentation of 80 abdominal MRI sets along with their ground truth contours. The performance of the ACR process was evaluated using 4 contour accuracy metrics: the Dice similarity coefficient (DSC), mean distance to agreement (MDA), surface DSC, and added path length (APL) on the auto-segmented contours of the small bowel, large bowel, combined bowels, pancreas, duodenum, and stomach. Results A portion (3%-39%) of the corrected contours became practically acceptable per the American Association of Physicists in Medicine Task Group 132 (TG-132) recommendation (DSC >0.8 and MDA <3 mm). The best correction performance was seen in the combined bowels, where for the contours with major errors (initial DSC <0.5 or MDA >8 mm), the mean DSC increased from 0.34 to 0.59, the mean MDA decreased from 7.02 mm to 5.23 mm, and the APL reduced by almost 20 mm, whereas for the contours with minor errors, the mean DSC increased from 0.72 to 0.79, the mean MDA decreased from 3.35 mm to 3.29 mm, and more than one-third (39%) of the ACR contours became clinically acceptable. The execution time for the ACR process on one subregion was less than 2 seconds using an NVIDIA GTX 1060 GPU. Conclusions The ACR process implemented based on the ACM was able to quickly correct for some inaccurate contours produced from MRI-based deep learning auto-segmentation of complex abdominal anatomy. The ACR method may be integrated into the auto-segmentation step to accelerate the process of MRI-guided ART.

Dataset Information

Feasibility of Continual Deep Learning-Based Segmentation for Personalized Adaptive Radiation Therapy in Head and Neck Area.

Publications

Feasibility of Continual Deep Learning-Based Segmentation for Personalized Adaptive Radiation Therapy in Head and Neck Area.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets