Deep learning for autosegmentation for radiotherapy treatment planning: State-of-the-art and novel perspectives

Strahlentherapie und Onkologie, Journal Year: 2024, Volume and Issue: unknown

Published: Aug. 6, 2024

Abstract The rapid development of artificial intelligence (AI) has gained importance, with many tools already entering our daily lives. medical field radiation oncology is also subject to this development, AI all steps the patient journey. In review article, we summarize contemporary techniques and explore clinical applications AI-based automated segmentation models in radiotherapy planning, focusing on delineation organs at risk (OARs), gross tumor volume (GTV), target (CTV). Emphasizing need for precise individualized plans, various commercial freeware state-of-the-art approaches. Through own findings based literature, demonstrate improved efficiency consistency as well time savings different scenarios. Despite challenges implementation such domain shifts, potential benefits personalized treatment planning are substantial. integration mathematical growth detection further enhances possibilities refining volumes. As advancements continue, prospect one-stop-shop represents an exciting frontier radiotherapy, potentially enabling fast enhanced precision individualization.

Language: Английский

---

NRG Oncology Assessment of Artificial Intelligence Deep Learning–Based Auto-segmentation for Radiation Therapy: Current Developments, Clinical Considerations, and Future Directions

International Journal of Radiation Oncology*Biology*Physics, Journal Year: 2023, Volume and Issue: 119(1), P. 261 - 280

Published: Nov. 14, 2023

Language: Английский

---

Generation and evaluation of anatomy‐preserving virtual CT for online adaptive proton therapy

Medical Physics, Journal Year: 2024, Volume and Issue: 51(3), P. 1536 - 1546

Published: Jan. 17, 2024

Abstract Background Daily CTs generated by CBCT correction are required for daily replanning in online‐adaptive proton therapy (APT) to effectively deal with inter‐fractional changes. Out of the currently available methods, suitability a CT generation method dose calculation also depends on anatomical site. Purpose We propose an anatomy‐preserving virtual (APvCT) as hybrid correction, which is especially suitable large anatomy deformations. The accuracy was assessed comparison corrected (cCBCT) and (vCT) methods context online APT. Methods Seventy‐one CBCTs four prostate cancer patients treated intensity modulated (IMPT) were converted using cCBCT, vCT, newly proposed APvCT method. In APvCT, planning (pCT) mapped geometry deformable image registration boundary conditions controlling regions interest (ROIs) created deep learning segmentation cCBCT. relative frequency distribution (RFD) HU, mass density stopping power ratio (SPR) values compared pCT. ROIs vCT cCBCT terms Dice similarity coefficient (DSC) mean distance‐to‐agreement (mDTA). For each patient, robustly optimized IMPT plan pCT subsequent adaptive plans CTs. same anatomy, recalculated corresponding APvCT. distributions isodose volumes 3D global gamma‐index passing rate (GPR) at γ(2%, 2 mm) criterion. Results all patients, no noticeable difference RFDs observed amongst except showed difference. minimum DSC value 0.96 0.39 contours respectively. average mDTA 0.01 cm clinical target volume ≤0.01 organs risk, increased 0.18 ≤0.52 vCT. GPR 90.9%, 64.5%, 67.0% versus When resulted GPRs 89.5 ± 5.1% 65.9 19.1%, 80.0%, 90.0%, 95.0%, 98.0%, 100.0% 0.97, 0.95, 0.91 plans, 0.89, 0.88, 0.87, 0.85, 0.81 plans. Hausdorff distance some cases exceeded 1.00 cm. Conclusions good agreement reference indicates preservation A erroneous can result incorrect plan. Further, slightly lower between cCBCT‐based be explained cCBCT's SPR RFD from

Language: Английский

---

Fully automated radiotherapy treatment planning: A scan to plan challenge

Radiotherapy and Oncology, Journal Year: 2024, Volume and Issue: 200, P. 110513 - 110513

Published: Sept. 1, 2024

Language: Английский

---

Deep learning for automated segmentation in radiotherapy: a narrative review

British Journal of Radiology, Journal Year: 2023, Volume and Issue: 97(1153), P. 13 - 20

Published: Dec. 12, 2023

The segmentation of organs and structures is a critical component radiation therapy planning, with manual being laborious time-consuming task. Interobserver variability can also impact the outcomes therapy. Deep neural networks have recently gained attention for their ability to automate tasks, convolutional (CNNs) popular approach. This article provides descriptive review literature on deep learning (DL) techniques in planning. focuses five clinical sub-sites finds that U-net most commonly used CNN architecture. studies using DL image were included brain, head neck, lung, abdominal, pelvic cancers. majority articles planning concentrated normal tissue structures. N-fold cross-validation was employed, without external validation. research area expanding quickly, standardization metrics independent validation are benchmarking comparing proposed methods.

Language: Английский

---

The impact of multicentric datasets for the automated tumor delineation in primary prostate cancer using convolutional neural networks on 18F-PSMA-1007 PET

Radiation Oncology, Journal Year: 2024, Volume and Issue: 19(1)

Published: Aug. 7, 2024

Abstract Purpose Convolutional Neural Networks (CNNs) have emerged as transformative tools in the field of radiation oncology, significantly advancing precision contouring practices. However, adaptability these algorithms across diverse scanners, institutions, and imaging protocols remains a considerable obstacle. This study aims to investigate effects incorporating institution-specific datasets into training regimen CNNs assess their generalization ability real-world clinical environments. Focusing on data-centric analysis, influence varying multi- single center approaches algorithm performance is conducted. Methods nnU-Net trained using dataset comprising 161 18 F-PSMA-1007 PET images collected from four distinct institutions (Freiburg: n = 96, Munich: 19, Cyprus: 32, Dresden: 14). The partitioned such that data each are systematically excluded used solely for testing model's generalizability unfamiliar sources. Performance compared through 5-Fold Cross-Validation, providing detailed comparison between models centers those aggregated multi-center datasets. Dice Similarity Score, Hausdorff distance volumetric analysis primary evaluation metrics. Results mixed approach yielded median DSC 0.76 (IQR: 0.64–0.84) five-fold cross-validation, showing no significant differences (p 0.18) with exclusion center, which performed 0.74 0.56–0.86). Significant improvements regarding were observed Dresden cohort (multi-center 0.71, IQR: 0.58–0.80 vs. single-center 0.68, 0.50–0.80, p < 0.001) Cyprus 0.74, 0.62–0.83 0.72, 0.54–0.82, 0.01). While Munich Freiburg also showed training, results statistical significance (Munich: 0.60–0.80 0.59–0.82, > 0.05; Freiburg: 0.78, 0.53–0.87 0.53–0.83, 0.23). Conclusion auto intraprostatic GTV multiple mostly generalize well unseen other centers. Training multicentric can improve exclusively segmentation. segmentation same CNN vary depending employed testing.

Language: Английский

---

Safety and efficiency of a fully automatic workflow for auto-segmentation in radiotherapy using three commercially available deep learning-based applications

Physics and Imaging in Radiation Oncology, Journal Year: 2024, Volume and Issue: 31, P. 100627 - 100627

Published: July 1, 2024

Advancements in radiotherapy auto-segmentation necessitate reliable and efficient workflows. Therefore, a standardized fully automatic workflow was developed for three commercially available deep learning-based applications compared to manual safety efficiency. The underwent evaluation with failure mode effects analysis. Notably, eight modes were reduced, including seven severity factors ≥7, indicating the effect on patients, two Risk Priority Number value >125, which assesses relative risk level. Efficiency, measured by mouse clicks, showed zero clicks workflow. This automation illustrated improvement both efficiency of

Language: Английский

---

Guidance on selecting and evaluating AI auto-segmentation systems in clinical radiotherapy: insights from a six-vendor analysis

Physical and Engineering Sciences in Medicine, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 13, 2025

Abstract Artificial Intelligence (AI) based auto-segmentation has demonstrated numerous benefits to clinical radiotherapy workflows. However, the rapidly changing regulatory, research, and market environment presents challenges around selecting evaluating most suitable solution. To support adoption of AI systems, Selection Criteria recommendations were developed enable a holistic evaluation vendors, considering not only raw performance but associated risks uniquely related deployment AI. In-house experience key bodies work on ethics, standards, best practices for in Radiation Oncology reviewed inform selection criteria strategies. A retrospective analysis using was performed across six including quantitative assessment five metrics (Dice, Hausdorff Distance, Average Surface Dice, Added Path Length) 20 head neck, thoracic, 19 male pelvis patients models as March 2023. total 47 identified seven categories. showed that overall no vendor exceedingly well, with systematically poor Data Security & Responsibility, Vendor Support Tools, Transparency Ethics. In terms performance, vendors varied widely from excellent poor. As new regulations come into force scope systems adapt needs, continued interest ensuring safe, fair, transparent will persist. The framework provided herein aims promote user confidence by exploring breadth clinically relevant factors informed decision-making.

Language: Английский

---

Under-representation for Female Pelvis Cancers in Commercial Auto-segmentation Solutions and Open-source Imaging Datasets

Clinical Oncology, Journal Year: 2025, Volume and Issue: unknown, P. 103651 - 103651

Published: Jan. 1, 2025

Language: Английский

---

Evaluation and failure analysis of four commercial deep learning‐based autosegmentation software for abdominal organs at risk

Journal of Applied Clinical Medical Physics, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 13, 2025

Abstract Purpose Deep learning‐based segmentation of organs‐at‐risk (OAR) is emerging to become mainstream in clinical practice because the superior performance over atlas and model‐based autocontouring methods. While several commercial deep autosegmentation solutions are now available, implementation these tools still at such a primitive stage that acceptance criteria underdeveloped due lack knowledge about systems’ tendencies failure modes. As starting point iterative process implementation, this study focuses on outlier analysis four for abdominal OARs. Materials methods The software, developed by Limbus AI, MIM Contour ProtégéAI, Radformation AutoContour, Siemens syngo.via, were used segment 111 patient cases. Geometric accuracy was quantitatively compared with contours using dice similarity coefficient (DSC) 95% Hausdorff distance (HD95). outliers from quantitative evaluations each software analyzed liver, stomach, kidneys possible causes summarized into six categories: (1) difference contouring style or guideline, (2) image acquisition quality, (3) abnormal anatomy OAR, (4) abutting organs/tissues, (5) external/internal devices, (6) other causes. Results For liver segmentation, most prominent cause discrepancies Limbus, which occurred its outliers, existence biliary stent internal/external drain as well resulting pneumobilia. included organs shared CT numbers similar those 5/8 outliers. 12 13 Radformation's heart and/or stomach while not only presence barium 5/11 but also produced fragmented Only provided imaging contrast directly caused incomplete delineation 10/12 21/25 kidneys, consistently followed RTOG guidelines, whereas institutional excluded renal pelvis some cases, 19/25 18/23 By contrast, appeared follow different guidelines exclude pelvis. Fragmented kidney found 10/15 25/26 ones linked use IV imaging, there enough evidence identify origin Limbus's contours. Conclusion OAR. This work can help vendors improve their inform users potential modes when tools.

Language: Английский

---