Overview of the Head and Neck Tumor Segmentation for Magnetic Resonance Guided Applications (HNTS-MRG) 2024 Challenge

Lecture notes in computer science, Journal Year: 2025, Volume and Issue: unknown, P. 1 - 35

Published: Jan. 1, 2025

Language: Английский

---

Assessment of contour accuracy in head and neck replanning: Deep learning trained model compared with deformable image registration propagation technique

Medical dosimetry, Journal Year: 2025, Volume and Issue: unknown

Published: May 1, 2025

Accurate contouring is crucial for optimal treatment outcomes, whether nonadaptive radiotherapy with single images or adaptive (ART) multiple images. For ART there are 2 common approaches automated segmentation: deformable image registration (DIR) propagation of prior contours from a previous to newer replanning (ii) deep learning (DL) generated by models trained datasets. The accuracy the latter approach impacted size, diversity and quality training dataset while former depends on contours, contrast between pairs, DIR algorithm used. This study assesses commercially available pretrained DL model (Mirada DLC04, DLC13, DLC14) tools (Velocity, MIM, Eclipse) generating in scenarios head neck region. Datasets region (n = 9 patients) included CTs 18) clinically approved doses. Manual contour data were compared against propagated (rigid, Velocity, Eclipse). Evaluation involved (a) clinical relevance scores, (b) grading (c) assessment manual style Radiation Oncologist Consultant Brouwer guideline (d) based geometric dosimetric metrics, These metrics dice similarity coefficient (DSC), mean distance agreement (MDA), Hausdorff distance(HD), volume ratio, dose ratio. Contours shortlisted statistical analysis (i) scores (iii) existing data. Statistical assessed Velocity as comparator. Contour relevancy highest spinal cord, parotids, oral cavity, mandible, larynx, brainstem. indicated most acceptable minor edits both except brachial plexus cavity variation described Oncologist. brainstem parotid analysis, indicating that: no evidence (all p > 0.1) difference contours; geometrically, (Velocity) was superior DLCExpert) terms MDA (p 0.014) HD < 0.001) 0.045); parotids amongst rigid registrations. In our region, DIR-based demonstrated glands DLCexpert). Among algorithms, significant differences observed, MIM ratio Our found among methods.

Language: Английский

---

Analyzing the Relationship between Dose and Geometric Agreement Metrics for Auto-Contouring in Head and Neck Normal Tissues

Diagnostics, Journal Year: 2024, Volume and Issue: 14(15), P. 1632 - 1632

Published: July 29, 2024

This study aimed to determine the relationship between geometric and dosimetric agreement metrics in head neck (H&N) cancer radiotherapy plans. A total 287 plans were retrospectively analyzed, comparing auto-contoured clinically used contours using a Dice similarity coefficient (DSC), surface DSC (sDSC), Hausdorff distance (HD). Organs-at-risk (OARs) with ≥200 cGy dose differences from clinical contour terms of D

Language: Английский

---

Artificial Intelligence Uncertainty Quantification in Radiotherapy Applications - A Scoping Review

medRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: May 13, 2024

Abstract Background/purpose The use of artificial intelligence (AI) in radiotherapy (RT) is expanding rapidly. However, there exists a notable lack clinician trust AI models, underscoring the need for effective uncertainty quantification (UQ) methods. purpose this study was to scope existing literature related UQ RT, identify areas improvement, and determine future directions. Methods We followed PRISMA-ScR scoping review reporting guidelines. utilized population (human cancer patients), concept (utilization UQ), context (radiotherapy applications) framework structure our search screening process. conducted systematic spanning seven databases, supplemented by manual curation, up January 2024. Our yielded total 8980 articles initial review. Manuscript data extraction performed Covidence. Data categories included general characteristics, RT characteristics. Results identified 56 published from 2015-2024. 10 domains applications were represented; most studies evaluated auto-contouring (50%), image-synthesis (13%), multiple simultaneously (11%). 12 disease sites represented, with head neck being common site independent application space (32%). Imaging used 91% studies, while only 13% incorporated dose information. Most focused on failure detection as main (60%), Monte Carlo dropout commonly implemented method (32%) ensembling (16%). 55% did not share code or datasets. Conclusion revealed diversity beyond auto-contouring. Moreover, clear additional methods, such conformal prediction. results may incentivize development guidelines implementation RT.

Language: Английский

---

Global Workforce and Access: Demand, Education, Quality

Seminars in Radiation Oncology, Journal Year: 2024, Volume and Issue: 34(4), P. 477 - 493

Published: Sept. 11, 2024

Language: Английский

---