Добавил:
Sekretar
kiopkiopkiop18@yandex.ru
t.me/Prokururor I Вовсе не секретарь, но почту проверяю
Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз:
Предмет:
Файл:Ординатура / Хирургия / Библиотека им академика М.И. Перельмана / Книга_5525_Библиотеки_им_академика_М_И_Перельмана.pdf
X
- •Foreword
- •Acknowledgments
- •Editor biographies
- •Yi Wang
- •X. Sharon Qi
- •List of contributors
- •1.2.3 Feature engineering and representation
- •1.2.4 Linear separability
- •1.2.5 Classical models
- •1.1 A brief introduction to AI
- •1.2 Machine learning basics
- •1.2.1 Learning paradigms
- •1.3 Artificial neural networks
- •1.3.1 Feed-forward neural networks
- •1.3.2 Recurrent neural networks
- •1.3.3 Convolutional neural networks
- •1.3.4 Attention
- •1.3.5 Training neural networks
- •1.3.6 Applications and use cases of deep learning
- •1.4 Model training and evaluation
- •1.4.1 Hyperparameters
- •1.4.2 Data split
- •1.4.3 Evaluation metrics
- •1.5 Generative models
- •1.5.1 Generative adversarial networks
- •1.5.2 Diffusion models
- •1.5.3 Applications and use cases
- •1.6 Ethical consideration and bias
- •1.6.1 Transparency and explainability
- •1.6.2 Bias and fairness
- •1.6.3 Data privacy violation
- •1.6.4 Risk and misuse
- •1.7 Summary
- •References
- •2.1 Introduction
- •2.1.2 Staff roles in radiation therapy
- •2.2 Overview of AI in radiation therapy
- •2.2.1 Patient evaluation and dose prescription
- •2.2.2 Treatment simulation
- •2.2.3 Contouring
- •2.2.4 Treatment planning
- •2.2.5 Quality assurance
- •2.2.6 Treatment delivery
- •2.2.7 Response assessment and toxicity management
- •2.3 Summary
- •3.1 Introduction
- •3.1.1 Introduction of clinical decision making and AI
- •3.1.2 The role of AI in clinical decision making
- •3.2 AI algorithms for clinical decision making
- •3.2.2 Radiomics
- •3.2.3 Data integration by AI
- •3.2.4 Interpretability of AI models
- •3.3 Application of AI in clinical decision making
- •3.3.1 Diagnosis and disease phenotyping
- •3.3.2 Personalized treatment
- •3.3.3 Treatment outcome and prognosis prediction
- •3.4 Challenges and future directions of AI in clinical decision making
- •3.4.1 Challenges and concerns
- •3.4.2 Future directions
- •3.5 Summary
- •References
- •4.1 Introduction
- •4.2 Imaging for treatment planning
- •4.2.1 CT simulation
- •4.2.2 4D-CT
- •4.2.3 PET/CT
- •4.2.4 MRI
- •4.3 Imaging for treatment guidance
- •4.3.1 Portal imaging
- •4.3.2 CBCT
- •4.3.3 CT-on-rail and CT-linac
- •4.3.4 MR-linac
- •4.3.5 PET-linac
- •4.4 Imaging for motion management
- •4.4.1 ExacTrac
- •4.4.2 Varian triggered imaging
- •4.4.3 4D-CBCT
- •4.4.4 Cine MRI
- •4.4.5 4D-MRI
- •4.4.6 Surface imaging
- •4.5 Imaging for treatment assessment
- •4.5.1 Contrasted CT
- •4.5.2 PET/CT
- •4.5.3 Functional MRI
- •4.6 Summary
- •5.1 Introduction to big data in radiation oncology
- •5.1.1 Overview of big data
- •5.1.2 Sources of big data in radiation oncology
- •5.1.3 Big data and AI in radiation oncology
- •5.2 Big data lifecycle in radiation oncology
- •5.2.1 Data aggregation and storage
- •5.2.2 Data sharing and security
- •5.4 The application of big data in radiation oncology
- •5.4.1 Medical image segmentation
- •5.4.2 Automatic treatment planning
- •5.4.3 Treatment response prediction
- •5.4.4 Quality assurance and patient safety
- •5.4.5 Clinical decision support
- •5.2.3 Data visualization
- •5.2.4 Knowledge creation and implementation
- •5.2.5 Data archiving and deletion
- •5.3 Big data analytics with AI
- •5.3.1 Data processing and integration
- •5.3.2 AI modeling
- •5.5 Challenges and future perspectives
- •5.6 Summary
- •Reference
- •6.1 The road to ART
- •6.1.1 3D conformal radiotherapy (3DCRT)
- •6.1.2 Intensity modulated radiotherapy (IMRT)
- •6.1.3 Image-guided radiotherapy (IGRT)
- •6.1.4 Adaptive radiotherapy (ART)
- •6.2 ART workflow and implementation
- •6.2.2 Current practice
- •6.2.3 Clinical impact
- •6.3 Considerations for implementing online ART
- •6.3.1 Time as a limiting factor
- •6.3.2 Implications for fast and reliable re-planning
- •6.3.4 Clinical considerations
- •6.4 Summary
- •7.1 Components of ART workflow
- •7.1.1 Simulation
- •7.1.2 Pre-planning
- •7.1.3 Online imaging and daily re-planning
- •7.1.4 Quality assurance
- •7.2 AI-driven ART
- •7.2.1 Simulation
- •7.2.2 Pre-planning
- •7.2.3 AI for delivery
- •7.3 Outlook and future directions
- •7.3.1 Real-time ART with AI
- •7.3.2 Dose escalation and functional adaption with AI
- •7.4 Summary
- •References
- •8.1 Introduction
- •8.2 Synthetic CT: deep learning methods
- •8.2.1 Conventional methods
- •8.2.2 U-Net
- •8.2.3 Generative adversarial networks
- •8.2.4 Denoising diffusion probabilistic model
- •8.3 Synthetic CT from CBCT
- •8.3.1 Noise and artifact reduction
- •8.3.2 Online dose calculation
- •8.3.3 Online image segmentation
- •8.4 Synthetic CT from MRI
- •8.4.1 Synthetic image accuracy
- •8.4.2 Dose calculation in MR-only radiation therapy
- •8.4.3 PET attenuation correction
- •8.4.4 Image registration
- •8.5 Discussion and outlook
- •8.6 Summary
- •References
- •9.1 AI-based image registration and segmentation for ART
- •9.1.1 Adaptive radiation therapy
- •9.2 Artificial intelligence
- •9.2.1 What is machine learning?
- •9.2.2 What is deep learning?
- •9.3 Deep learning: the basic components
- •9.3.1 Convolutional neural networks: looking at the picture
- •9.3.2 Pooling layers: keeping what matters most
- •9.3.3 Fully connected (dense) layers: bringing it all together
- •9.3.4 Activations
- •9.3.5 Loss: driving the model
- •9.3.6 Auto-encoders: remove the noise
- •9.3.7 Supervised versus unsupervised learning
- •9.3.8 Pre-trained convolutional neural networks
- •9.4 Image registration: bringing two images together
- •9.4.1 Registration similarity metrics
- •9.4.2 Types of registrations
- •9.5 AI-based image registration
- •9.5.1 Supervised learning
- •9.5.2 Unsupervised learning
- •9.5.3 Registration in ART
- •9.5.4 Commonalities in architectures
- •9.6 Image segmentation
- •9.6.1 Introduction: coloring by the numbers
- •9.6.2 Segmentation networks
- •9.6.3 Best practices
- •9.7 Summary
- •10.1 Introduction
- •10.1.1 Overview of chapter content
- •10.2 The landscape of AI-assisted dose prediction
- •10.2.1 Traditional machine learning for dose prediction
- •10.2.2 Deep learning-based dose prediction
- •10.2.3 Challenges in AI-assisted dose prediction
- •10.3 Re-planning workflows powered by AI
- •10.3.1 Deep learning for re-planning pipelines
- •10.4 Future directions of AI-assisted dose prediction and re-planning
- •10.5 Summary
- •11.1 Introduction
- •11.2.1 Imaging-based motion monitoring
- •11.2.2 Delivery system actions
- •11.2.3 Challenges for real-time ART implementation
- •11.3 AI in real-time ART workflows
- •11.3.1 Improving intrafraction motion monitoring through AI
- •11.3.2 Mitigating system latency through AI
- •11.4 AI for ART delivery: future directions
- •11.4.1 Management of non-respiratory motion
- •11.4.2 Training AI models with small or unpaired datasets
- •11.4.4 Biology-guided ART delivery
- •11.5 Summary
- •References
- •12.1 Introduction
- •12.2 Patient QA
- •12.2.1 Pre-planning QA
- •12.2.2 Pre-treatment plan QA
- •12.2.3 On-treatment QA
- •12.3 Treatment delivery systems and instruments
- •12.3.1 Machine commissioning
- •12.3.2 Machine QA
- •12.3.3 Dosimetry tool QA
- •12.4 Summary
- •References
- •13.1 Data resources for response modeling in radiotherapy
- •13.1.1 Clinical data
- •13.1.2 Imaging (radiomics)
- •13.1.3 Treatment planning (dosiomics)
- •13.1.4 Multiomics
- •13.2 Radiotherapy treatment outcome modeling
- •13.2.1 TCP/NTCP in radiotherapy
- •13.2.2 Clinical outcomes versus PROs
- •13.2.3 Machine learning response prediction
- •13.2.4 Explainability of ML response models
- •13.2.5 Sample use cases
- •13.3 AI response-based adaptive radiotherapy
- •13.3.1 Requirements and challenges
- •13.3.2 Prediction versus treatment optimization
- •13.3.3 Sample use cases
- •13.4 Challenges and recommendations
- •13.5 Summary
- •Acknowledgments
- •References
- •14.1 Overview of challenges in AI-driven ART
- •14.2 Data challenges
- •14.2.1 Data availability
- •14.2.2 Data quality
- •14.2.3 Data privacy
- •14.3 Technical challenges
- •14.3.2 Model robustness and generalizability
- •14.3.3 Model explainability and interpretability
- •14.4 Challenges associated with online and real-time workflows
- •14.4.1 Image quality
- •14.4.2 Dose calculation
- •14.4.3 Real-time ART
- •14.5 Operational challenges
- •14.5.1 Clinical validation
- •14.5.3 Staff training
- •14.5.4 User experiences
- •14.5.5 Quality management program
- •14.5.6 Financial challenges
- •14.6 Ethical, regulatory, and legal challenges
- •14.6.1 Ethical issues
- •14.6.2 Regulatory and legal issues
- •14.7 Summary
- •References
- •15.1 Clinical considerations for CT-based offline ART
- •15.1.1 Patient and site selection
- •15.1.2 Re-simulation
- •15.1.3 Re-planning
- •15.1.4 Plan summation and evaluation
- •15.1.6 Limitations and future directions
- •15.2 Clinical considerations for CBCT/CT-based online ART
- •15.2.2 Patient and site selection
- •15.2.3 Simulation
- •15.2.4 Pre-planning review
- •15.2.5 Reference planning
- •15.2.9 Limitations and future directions
- •15.3 Summary
- •References
- •16.1 Introduction
- •16.2 Overview of MRI-guided ART systems
- •16.3 MRI-guided ART workflow
- •16.4 AI applications for MRI-guided ART
- •16.4.1 Synthetic CT generation
- •References
- •16.4.2 Auto-segmentation
- •16.4.3 Image registration
- •16.4.4 Others
- •16.4.5 Future AI development and implementation
- •16.5 Summary
- •17.1 Functional PET-guided ART
- •17.1.1 PET-based functional imaging overview
- •17.1.2 From anatomy to function: the power of PET in radiation therapy
- •17.1.5 Conclusions and future prospects
- •17.2 Functional MRI-guided ART
- •17.2.1 From anatomy to function: the power of functional MRI in radiation therapy
- •17.2.4 Conclusion and future prospects
- •17.3 Summary
- •References
- •18.1 Proton ART
- •18.1.1 Clinical context and necessity
- •18.1.2 Patient populations
- •18.1.4 Rationale for AI in proton ART
- •18.2 AI in proton ART
- •18.2.1 Imaging
- •18.2.2 Deformable and rigid registration
- •18.2.3 Contour propagation
- •18.2.4 Dose calculations
- •18.2.5 Plan optimization
- •18.2.6 Other developments
- •18.3 Implementation of adaptive proton therapy
- •18.4 Summary
- •References
- •19.1 Designing clinical trials with AI
- •19.1.1 The essential role of clinical trials
- •19.1.2 Trial protocols and methodologies
- •19.1.3 AI-driven clinical trial design and execution
- •19.1.4 Incorporation of digital twins (DTs) in clinical trials
- •19.2 Implementation of AI in ongoing clinical trials
- •19.2.1 Integration with existing clinical trial frameworks
- •19.2.2 Quality assurance, compliance, and standardization
- •19.3 Case studies of AI in adaptive radiotherapy trials
- •19.3.1 Overview of guidance for advanced radiotherapy in clinical trials
- •19.3.2 AI in the radiotherapy clinical trial quality assurance processes
- •19.4 Ethical and regulatory considerations
- •19.4.1 Patient consent and data privacy
- •19.4.2 Bias, fairness, and transparency
- •19.4.3 Regulatory guidelines and compliance
- •19.5 Future directions and challenges
- •19.5.1 Emerging technologies and techniques
- •19.5.2 Alternative strategies
- •19.6 Conclusion
- •19.7 Summary
- •References
- •20.1 Risk management
- •20.1.1 Prospective risk assessments
- •20.1.2 Root cause analysis

Artificial Intelligence in Adaptive Radiation Therapy
[161] Fuhrman J D, Gorre N, Hu Q, Li H, El Naqa I and Giger M L 2022 A review of
explainable and interpretable AI with applications in COVID-19 imaging Med. Phys.
1–14
[162] Ribeiro M T, Singh S and Guestrin C 2016 Why should I trust you?’ Explaining the
predictions of any classifier Proc. of the 22nd ACM SIGKDD Int. Conf. on Knowledge
Discovery and Data Mining pp 1135–44
[163] Lundberg S M and Lee S-I 2017 A unified approach to interpreting model predictions Proc.
31st Int. Conf. on Neural Information Processing SystemsPages pp 4768–77
[164] Selvaraju R R, Cogswell M, Das A, Vedantam R, Parikh D and Batra D 2017 Grad-cam:
Visual explanations from deep networks via gradient-based localization Proc. of the IEEE
Int. Conf. on Computer Vision pp 618–26
[165] Saednia K et al 2020 Quantitative thermal imaging biomarkers to detect acute skin toxicity
from breast radiation therapy using supervised machine learning Int. J. Radiat. Oncol. Biol.
Phys.
106 1071–83
[166] Diamant A, Chatterjee A, Vallières M, Shenouda G and Seuntjens J 2019 Deep learning in
head and neck cancer outcome prediction Sci. Rep.
[167] Nayan M et al 2022 A machine learning approach to predict progression on active
surveillance for prostate cancer Urol. Oncol. Semin. Ori. Investig.
[168] Wei L et al 2021 A deep survival interpretable radiomics model of hepatocellular carcinoma
patients Phys. Med. Eur. J. Med. Phys.
[169] Katzman J L, Shaham U, Cloninger A, Bates J, Jiang T and Kluger Y 2018 DeepSurv:
personalized treatment recommender system using a Cox proportional hazards deep neural
network BMC Med. Res. Methodol.
[170] Yan D, Vicini F, Wong J and Martinez A 1997 Adaptive radiation therapy Phys. Med.
Biol.
42 123
[171] Tseng H-H, Luo Y, Ten Haken R K and El Naqa I 2018 The role of machine learning in
knowledge-based response-adapted radiotherapy Front. Oncol.
[172] Tseng H-H, Ten Haken R K and El Naqa I 2022 Smart adaptive treatment strategies
Machine and Deep Learning in Oncology, Medical Physics and Radiology
M J Murphy (Cham: Springer International) pp 439–52
[173] Ger R B, Wei L, El Naqa I and Wang J 2023 The promise and future of radiomics for
personalized radiotherapy dosing and adaptation Semin. Radiat. Oncol.
[174] Niraula D, Jamaluddin J, Matuszak M M, Haken R K T and El Naqa I 2021 Quantum
deep reinforcement learning for clinical decision support in oncology: application to
adaptive radiotherapy Sci. Rep.
[175] Niraula D et al 2023 A clinical decision support system for AI-assisted decision-making in
response-adaptive radiotherapy (ARCliDS) Sci. Rep.
[176] Sun W et al 2022 Precision radiotherapy via information integration of expert human
knowledge and AI recommendation to optimize clinical decision making Comput. Methods
Prog. Biomed.
[177] Tseng H-H, Luo Y, Cui S, Chien J-T, Ten Haken R K and El Naqa I 2017 Deep
reinforcement learning for automated radiation adaptation in lung cancer Med. Phys.
6690–705
[178] Luo Y et al 2018 A multiobjective Bayesian networks approach for joint prediction of
tumor local control and radiation pneumonitis in nonsmall-cell lung cancer (NSCLC) for
response-adapted radiotherapy Med. Phys.
221 106927
11 23545
82 295–305
18 24
45 3980–95
9 2764
40 161.e1–7
8 266
ed I El Naqa and
33 252–61
13 5279
49
44
13-36

Artificial Intelligence in Adaptive Radiation Therapy
[179] Barto A G and Sutton R S 1998 Reinforcement Learning: An Introduction (Adaptive
Computation and Machine Learning) (Cambridge, MA: MIT Press)
[180] Green O L, Henke L E and Hugo G D 2019 Practical clinical workflows for online and
offline adaptive radiation therapy Semin. Radiat. Oncol.
[181] Yu T et al 2019 Pretreatment prediction of adaptive radiation therapy eligibility using
MRI-based radiomics for advanced nasopharyngeal carcinoma patients Front. Oncol.
1050
[182] Lam S-K et al 2022 Multi-organ omics-based prediction for adaptive radiation therapy
eligibility in nasopharyngeal carcinoma patients undergoing concurrent chemoradiotherapy
Front. Oncol.
[183] Zhang R, Cai Z, Luo Y, Wang Z and Wang W 2022 Preliminary exploration of response
the course of radiotherapy for stage III non-small cell lung cancer based on longitudinal CT
radiomics features Eur. J. Radiol. Open
[184] Forouzannezhad P et al 2022 Multitask learning radiomics on longitudinal imaging to
predict survival outcomes following risk-adaptive chemoradiation for non-small cell lung
cancer Cancers
[185] Lou B et al 2019 An image-based deep learning framework for individualising radiotherapy
dose: a retrospective analysis of outcome prediction Lancet Digit. Health
[186] Mierzwa M L et al 2022 Randomized phase II study of physiologic MRI-directed adaptive
radiation boost in poor prognosis head and neck cancer Clin. Cancer Res.
[187] Kong F-M et al 2017 Effect of midtreatment PET/CT-adapted radiation therapy with
concurrent chemotherapy in patients with locally advanced non-small-cell lung cancer: a
phase 2 clinical trial JAMA Oncol.
[188] Gameng H A, Gerardo B B and Medina R P 2019 Modified adaptive synthetic SMOTE to
improve classification performance in imbalanced datasets 2019 IEEE 6th Int. Conf. on
Engineering Technologies and Applied Sciences (ICETAS) pp 1–5
[189] Sharma A, Singh P K and Chandra R 2022 SMOTified-GAN for class imbalanced pattern
classification problems IEEE Access
[190] Chawla N V, Bowyer K W, Hall L O and Kegelmeyer W P 2002 SMOTE: synthetic
minority over-sampling technique J. Artif. Intell. Res.
[191] Dablain D, Krawczyk B and Chawla N V 2023 DeepSMOTE: fusing deep learning and
SMOTE for imbalanced data IEEE Trans. Neural Netw. Learn. Syst.
[192] He H, Bai Y, Garcia E A and Li S 2008 ADASYN: adaptive synthetic sampling approach
for imbalanced learning 2008 IEEE Int. Joint Conf. on Neural Networks (IEEE World
Congress on Computational Intelligence) (Piscataway, NJ: IEEE) pp 1322–8
[193] El Naqa I 2021 Prospective clinical deployment of machine learning in radiation oncology
Nat. Rev. Clin. Oncol.
11 792024
9 100391
14 1228
3 1358–65
10 30655–65
18 605–6
29 219–27
1 e136–47
28 5049–57
16 321–57
34 6390–404
9
13-37

IOP Publishing
Artificial Intelligence in Adaptive Radiation Therapy
Yi Wang and X. Sharon Qi
Chapter 14
Challenges of artificial intelligence
implementation in adaptive radiation therapy
Yi Wang and X. Sharon Qi
The integration of artificial intelligence (AI) into adaptive radiation therapy (ART)
holds immense promise for revolutionizing radiation therapy [1]. While the prior
chapters in this volume have thoroughly explored the use of AI in each step of the
ART workflow, this chapter carefully examines the multifaceted challenges accompanying the research, development, and clinical implementation of AI in ART.
Starting with a brief overview of the current landscape of AI-driven ART, the
chapter will discuss the challenges in various dimensions, including data, technical,
operational, ethical, regulatory, and financial [2, 3]. Developing reliable AI models
in clinical practice faces data-related challenges, such as limited availability, privacy
concerns, and imbalanced datasets. On the technical front, selecting optimal
architectures, managing transfer learning, estimating uncertainties, and ensuring
real-time performance add layers of complexity. Equally important is the human
factor—building clinician trust in AI recommendations, enhancing interpretability,
and training clinicians for AI collaboration demand careful attention. Additionally,
rigorous validation and evaluation processes, coupled with navigating regulatory
pathways, further complicate AI implementation. This chapter underscores the
critical need for model robustness, well-defined performance metrics, and strict
adherence to regulatory frameworks to facilitate seamless clinical adoption.
14.1 Overview of challenges in AI-driven ART
ART was first introduced as a closed-loop radiation treatment process that involves
modifying the initial treatment plan through systematic feedback from frequent
imaging acquisition, typically done on a daily basis [4]. The concept of ART has
evolved and emerged as a transformative approach in cancer treatment, allowing for
adjustments to treatment plans based on patient-specific anatomic, biological, and/
or functional changes during radiation therapy [1, 2]. ART, particularly the online
doi:10.1088/978-0-7503-6119-4ch14 14-1 ª IOP Publishing Ltd 2025. All rights,
including for text and data mining (TDM), artificial intelligence (AI) training, and similar technologies, are reserved.

Artificial Intelligence in Adaptive Radiation Therapy
version, has been widely implemented using various online image-guided radiation
therapy (IGRT) modalities, including computed tomography (CT) on-rail [5], conebeam CT (CBCT) [6], magnetic resonance imaging (MRI) [7], and positron emission
tomography (PET) [8]. These highly personalized approaches hold immense promise
for improving patient outcomes and minimizing side effects. As described in prior
chapters, the current landscape of AI in ART presents exciting possibilities [1, 3]. AI
algorithms are already demonstrating their power in various aspects of treatment,
from enhancing tumor segmentation and dose prediction to optimizing treatment
planning and assessing treatment response [1]. These advancements are driven by the
ability of AI to analyze vast amounts of medical data, including imaging, dosimetry,
and clinical records, to uncover hidden patterns and relationships that would elude
human analysis. However, integrating cutting-edge AI technologies into this
complex clinical workflow presents a multifaceted array of challenges that need to
be addressed before its full potential can be realized [9].
The path towards seamless AI integration in ART is paved with numerous
hurdles, which can be broadly be categorized into several key domains: data,
technical, operational, ethical, regulatory, and legal [1]. Moreover, online and realtime ART workflows induce additional challenges. In the following sections, each of
these challenges will be discussed in detail, with potential solutions explored. This
comprehensive review aims to facilitate safe implementation and smooth operation
of AI-ART, fulfilling its full potential to provide personalized radiation therapy.
14.2 Data challenges
One of the most substantial challenges on the path to AI-driven ART is data. This
section examines three key data obstacles on the development of robust AI models
for ART: data availability, quality, and privacy.
14.2.1 Data availability
While offline ART has many similarities with initial treatment planning, online and
real-time ART face additional challenges in data collection due to their unique
clinical workflow. Many current ART systems (e.g. Ethos by Varian Medical
Systems, Palo Alto, CA) operate as closed-loop environments, tightly integrating
data acquisition with adaptive treatment delivery [10]. This can make it difficult for
researchers and developers to access the raw data needed for AI training and
validation. The data may be locked within proprietary formats or require specific
permissions and authorization procedures. While the vendors and their authorized
researchers could gain access to the ART imaging and planning data, the obstacles
for independent researchers and developers to access these data could slow down AI
innovation in ART.
Despite the wide availability of the various ART platforms, e.g. the CBCT-based
Varian Ethos system [11 ], MRI-based ViewRay system [12], and Elekta Unity
system [13], the percentage of patients receiving ART treatment remains small
and the utilization of ART is limited to a few disease sites and protocols
14-2

Artificial Intelligence in Adaptive Radiation Therapy
(e.g. gastrointestinal tumors which are subject to daily variation of stomach and
bowel fillings, stereotactic body radiation therapy in which a large dose of radiation is
delivered in few fractions, and head and neck cancers which can grow or shrink during
the treatment course) [14–16]. Due to the lack of proper data, such as labeled target
and organs on fractional images (e.g. CBCT), specifically trained and evaluated AI
models are generally lacking or inaccurate for clinical ART applications. Despite the
efforts of bridging the two modalities with pseudo-CT to enable the use of the
CT-based segmentation model, the difference in imaging modality and image quality
adds additional uncertainty that may reduce the benefit of online adaptation [17]. With
further expansion of ART, more accurate AI models could be trained on intrinsic
ART data, eliminating the cross-modality uncertainty.
14.2.2 Data quality
Robust AI models rely on high-quality data. Building reliable and effective AI
models requires a robust foundation of accurate, balanced, diverse, and wellannotated data [18]. However, in the context of ART, this presents unique
challenges that impede the progress and effectiveness of AI implementation. ART
generates a dynamic stream of data, including imaging (e.g. CBCT), segmentation,
and dosimetry. The quality of these data might be inferior to those acquired for
initial simulation and planning [16]. The quality of the online images (e.g. CBCT)
which are used for adaptive re-planning often does not match that of the initial
simulation images (e.g. simulation CT), with the exception of the less-common,
space-extensive CT-on-rail system [19]. The time that the radiation oncologist or
ART team members spend on reviewing and editing the online contours is often
much shorter than what they spend during initial planning [20]. In some cases, the
online contours may be reviewed and edited by non-physician team members.
Finally, online ART requires a quick turn-around, and the capacity of the optimizer
of the online planning system (e.g. Varian Ethos) might not match that of its
counterpart for initial planning (e.g. Varian Eclipse) [21].
When developing AI models using data acquired in initial simulation and
treatment planning, it is generally possible to assemble a large and diverse dataset
while avoiding reliance on multiple datasets from the same patient. In contrast,
data acquired during online ART are far more limited, with models often trained
on fewer patients but multiple datasets from each adaptive fraction. As a result,
AI models trained by ART data are less likely to cover the broad diversity of
patient characters, such as the size, shape, and location of the tumor, as well as the
patient’s body size, gender, and race—factors that can affect the patient’s normal
anatomy [21].
14.2.3 Data privacy
The integration of AI technologies into ART presents substantial data privacy
challenges, requiring robust measures to safeguard sensitive patient information and
maintain trust. AI tools rely on extensive patient data—such as medical histories,
14-3

Artificial Intelligence in Adaptive Radiation Therapy
imaging studies, treatment plans, and outcomes—exposing this information to risks
such as unauthorized access, data breaches, and misuse [22]. Ensuring data
protection is paramount, demanding strict encryption methods, access controls,
and AI systems designed with inherent security measures to prevent cyberattacks
and ensure data integrity. Regular security updates address emerging vulnerabilities.
While anonymizing data for AI applications is crucial, risks of re-identification
persist, particularly when anonymized datasets are merged with public information.
Enhancing and validating anonymization techniques are essential for reducing
identification risks and complying with privacy regulations. Data sharing across
institutions, often required for AI model development, raises concerns about patient
consent and ethical compliance [18]. Transparent policies must be implemented to
inform patients about data usage, sharing, and protection, while explicit consent
fosters trust and fulfills legal and ethical obligations. Collaborations with external
vendors necessitate stringent data protection agreements and oversight, ensuring
adherence to high standards of privacy and security. Comprehensive vendor
management policies mitigate the risks associated with third-party involvement,
promoting consistent data protection practices.
14.3 Technical challenges
Beyond the data-related hurdles, this section will explore the technical challenges
associated with AI models used for ART, such as model accuracy, efficiency,
robustness, generalizability, explainability, interpretability, and computation speed.
14.3.1 Model accuracy and efficiency
Choosing the right AI model for a specific ART application is critical for the success
of this highly demanding technology [23]. While computation time is not usually
critical for AI applications in initial simulation and treatment planning, it becomes
mission-critical in online ART when the patient is on the treatment couch and in
real-time ART when multiple dynamic components of the treatment delivery system
are moving simultaneously. Each AI algorithm has its strengths and limitations in
different aspects such as efficiency, accuracy, robustness, and interpretability. For
online ART applications, AI models need to be highly efficient to minimize the time
between imaging and treatment. The longer the wait time, the more patient motion
occurs, reducing the benefits of adaptation. Conversely, since the ART team works
under significant time pressure, the AI models need to be highly robust to produce
reliable results, reducing the chance of human error or the need for compromise. A
prime example is dose prediction. In initial planning, it is generally acceptable for
the AI algorithm to complete dose prediction and plan optimization in 10–30 min, as
the treatment planner can multitask. However, the same process needs to be
completed in a few minutes in online ART and almost in real time in real-time
ART [20]. Additionally, in initial planning, the treatment planner, typically a
medical dosimetrist or physicist, has more time to identify, analyze, and troubleshoot any dose discrepancies. In online ART, however, the treatment planner, often
a medical dosimetrist or radiation physicist, is under significant time pressure,
14-4

Artificial Intelligence in Adaptive Radiation Therapy
leaving less time to create and evaluate the adaptive plan [1]. This constraint can
potentially compromise plan quality. Online ART follows a different treatment
planning workflow than the initial treatment plan, and it is desirable to have
specialized AI models to address specific challenges. Ideal solutions should optimize
both accuracy and efficiency to facilitate the time-critical decision-making processes,
such as contouring and adaptive re-planning [3].
14.3.2 Model robustness and generalizability Robustness refers to an AI model’s ability to maintain its performance and accuracy
when subjected to variations in input data or operating conditions [24]. In ART,
robustness is paramount due to significant variability in patients under treatment.
Factors such as differences in anatomy, tumor characteristics, imaging protocols
(e.g. kVp and mA settings in CBCT), equipment, and treatment protocols can all
impact AI model performance. A robust AI system in ART must be resilient to these
variations and capable of producing consistent results. Achieving robustness
requires thorough testing and validation of AI models using diverse datasets
encompassing a wide range of patient demographics, imaging techniques, and
clinical scenarios. This process helps identify potential weaknesses and refine the
models to enhance their resilience.
Generalizability is the ability of an AI model to apply its learned knowledge and
perform well on new, unseen data [25]. For ART, this means that an AI system
trained on a specific dataset should generalize its predictions and recommendations
to patients and clinical settings not included in the initial training data. One of the
primary challenges in achieving generalizability is ensuring that the training data are
representative of the broader patient population and clinical practices. If the training
data are biased or limited to a specific subset of patients or conditions, the AI model
may fail to generalize effectively, leading to inaccurate predictions and suboptimal
treatment recommendations for patients who differ from the training cohort [18]. To
enhance generalizability, AI models must be developed using diverse and representative datasets. Collaboration between multiple institutions and clinical sites can
help aggregate data from various sources, ensuring a more comprehensive training
dataset. Additionally, techniques such as transfer learning, which involves finetuning pre-trained models on new data, can improve the generalizability of AI
systems.
Robustness and generalizability are essential for the clinical success of AI in
ART. Achieving these attributes requires addressing challenges such as variability in
imaging protocols (e.g. kVp and mAs) and equipment (e.g. imagers), as well as
differences in patient populations (e.g. age, gender, weight, and medical history).
The continuous evolution of medical practices and technologies further necessitates
regular updates and validation of AI models. Key strategies to overcome these
hurdles include standardizing data collection and preprocessing protocols, diversifying training datasets, and leveraging advanced machine learning techniques to
enhance adaptability [25]. Rigorous validation across multiple clinical sites is vital to
identify and mitigate limitations, while clinician feedback and expert oversight
14-5

Artificial Intelligence in Adaptive Radiation Therapy
throughout development ensure that AI models meet the required standards.
Ensuring consistent performance across diverse conditions and generalization to
new data are critical for reliable ART delivery [18].
14.3.3 Model explainability and interpretability Explainability refers to the extent to which the internal workings of an AI model can
be understood by humans. It involves making the decision-making process of the
model transparent so that clinicians can comprehend why a particular recommendation or prediction is made. This is crucial for building trust, as clinicians need to
be confident in the AI’s outputs to rely on them in critical clinical settings [25].
Interpretability, on the other hand, is the degree to which a human can understand
the cause of a decision. It focuses on the clarity with which the AI’s predictions can
be presented and explained, making it easier for clinicians to grasp the reasoning
behind the AI model’s outputs. In the context of ART, especially in time-sensitive
situations, interpretability ensures that clinicians can quickly and accurately
interpret AI recommendations to make informed treatment decisions [25, 26].
Many advanced AI algorithms, such as deep learning neural networks
(DLNNs), involve numerous layers and parameters. They often operate as black
boxes, where the internal decision-making processes are not readily understandable even to experts. This makes it difficult to trace how specific inputs lead to
particular outputs. The lack of clear insights into how AI models arrive at their
conclusions can hinder the ART team’s abili ty to understand and trust the AI
model’s recommendations. Additionally, if the model’s outputs are not easily
interpretable, there is a risk of misinterpreting the recommendations, potentially
leading to errors in treatment decisions. Several strategies can be employed to
address these challenges. Designing user-friendly interfaces that present AI outputs
in an intuitive manner can facilitate quick interpretation. Interfaces that highlight
key information and offer clear, concise explanations enhance the clinicians’
ability to make informed decisions efficiently. Implementing visualization tools
that graphically represent the AI model’s reasoning can help make the internal
workings more transparent. Ongoing training and education on how to interpret
and utilize AI outputs effectively are also essential for users to understand why
specific decisions were made. Finally, collaborative efforts between AI developers
and model users can help improve performance, enhance transparency, and
promote trustworthiness [27].
14.3.4 Computational efficiency
The unique demands of ART, particularly in online and real-time settings, require
AI systems that deliver fast and accurate results without compromising performance
[28]. In online ART, where adaptive re-planning occurs while the patient remains on
the treatment couch, computation delays can increase patient motion risks and
reduce ART’s effectiveness. Real-time ART, involving continuous adaptation
during treatment delivery, demands even greater computational efficiency, as AI
models must provide immediate feedback to dynamic treatment components.
14-6

Artificial Intelligence in Adaptive Radiation Therapy
The complexity of AI models, with numerous layers and parameters, often limits
computational efficiency. Additionally, processing high-resolution imaging data and
performing online plan optimizations are computationally intensive, necessitating
significant hardware resources. Seamless integration with existing medical devices
and systems, as well as efficient data transfer pipelines, are essential to minimize
delays and ensure real-time performance. Optimizing AI algorithms, utilizing
specialized hardware, and streamlining clinical workflows are critical to achieving
the processing speeds required for online and real-time ART. Addressing these
challenges enables AI systems to deliver timely and accurate treatment adaptations,
unlocking ART’s full potential to improve treatment outcomes [18].
14.4 Challenges associated with online and real-time workflows
The primary benefit of online ART is its ability to optimize the dose distribution based
on inter-fractional variations such as daily changes in anatomy, biology, or function.
In real-time ART, dose distributions can be dynamically adjusted to account for intrafractional motion. However, both online and real-time ART workflows introduce new
uncertainties that may negate some of their intrinsic advantages. This section explores
the uncertainties arising from AI tools in these workflows.
14.4.1 Image quality
CBCT is the most common imaging modality used for online ART, as exemplified
by the Ethos system from Varian Medical Systems. However, the image quality of
CBCT is often inferior to that of simulation CT, posing significant challenges for
accurate contouring and dose calculations. To overcome these challenges, systems
such as Ethos utilize synthetic CT (sCT) images generated from CBCT data. This
approach allows deep learning (DL)-based deformable image registration (DIR) and
auto-segmentation models to generate autocontours on the sCT images. Despite
these advancements, synthetic CT cannot fully recover the lost contrast resolution
inherent in CBCT images, leading to inaccuracies in Hounsfield unit (HU) values
that may compromise the accuracy of online dose calculations. Additionally, current
sCT generation algorithms have limited capability in reducing the artifacts present
in the original CBCT images, such as metal artifacts or motion artifacts [29]. These
artifacts can propagate into the sCT images, affecting the quality of adaptive replanning. To fully realize the benefits of online ART, further research into hardware
improvements, advanced image reconstruction techniques, and artifact reduction
methods is necessary to enhance sCT image quality [30].
In MRI-based online ART, the superior soft tissue contrast of MRI allows for
more accurate contouring of tumors and organs compared to CT- or CBCT-guided
ART [7]. DL algorithms may struggle with regions that exhibit very low signal
intensities on MRI. For example, cortical bones have very low signal on MRI but
relatively high attenuation on CT, while metal implants may be invisible on MRI
but produce high attenuation and artifacts on CT. These discrepancies can lead to
mischaracterization of such regions during pseudo CT generation, causing uncertainty in HU values and subsequently affecting the accuracy of dose calculations.
14-7

Artificial Intelligence in Adaptive Radiation Therapy
Addressing these challenges necessitates the development of more sophisticated DL
models and training strategies capable of accurately representing these problematic
areas. Moreover, most MR simulators operate at high magnetic fields of 1.5 tesla (T)
or 3 T, such as SIGNA (GE Healthcare), MAGNETOM (Siemens Healthineers),
and Ingenia (Philips). While available online MR systems also operate at 1.5 T
(Unity from Elekta) [13], others function at much lower field strengths, such as the
0.35 T system (ViewRay from ViewRay Inc.) [12]. Due to the limited availability of
low-field MR data, DL-based auto-segmentation algorithms are often trained on
pre-treatment MRI acquired from high-field MR scanners. The inconsistency
between the training data (high-field strength) and the application (low-field
strength) can increase contouring uncertainty, hindering the full benefits for online
ART. Addressing this challenge requires either collecting low-field MR data for
training or developing DL models that are robust to variations in field strength.
14.4.2 Dose calculation
The use of AI, including machine learning (ML) and deep learning (DL), for dose
calculation in ART faces several challenges. One major issue is the computational
complexity of these ML and DL algorithms, which needs to conduct pixel-wise dose
prediction constrained by dose–volume histogram (DVH) requirements, within tight
time limits to enable online adaptation [31]. Variability in imaging protocols and
equipment across clinical settings can introduce inconsistencies in input data,
affecting the accuracy of dose predictions. Moreover, the diversity in patient
anatomy and tumor characteristics requires these models to generalize effectively
across different cases, which can be difficult without extensive and diverse training
datasets. Furthermore, ensuring the robustness against noise and artifacts in
imaging data is another critical challenge, as these factors can compromise dose
calculation accuracy. Additionally, integration with existing clinical workflows and
treatment delivery systems also poses difficulties, requiring seamless communication
between AI algorithms and hardware components. Currently, non-AI-based online
dose calculation approaches primarily rely on rapid re-optimization of the preadaptation plan (which could be the initial treatment plan or the most recent
fraction’s plan). For ML- or DL-based dose prediction algorithms, computation
speed must be as fast as, if not faster than, the current re-optimization approaches
[32]. Additional investigations are needed to find innovative methods to simplify
neural network structures to improve computational efficiency, while not compromising the accuracy of pixel and DVH-based dose predictions.
14.4.3 Real-time ART
Real-time ART holds the greatest potential for treating moving tumors that can be
visualized through real-time imaging, such as x-ray tracking and cine MRI [33]. In
real-time ART, the plan is continuously adapted based on real-time imaging,
requiring AI models to rapidly analyse new imaging data, predict future movement,
and adjust the radiation beam within milliseconds. This represents a paradigm shift
in cancer treatment, enabling dynamical adaptation to intra-fractional anatomical
14-8
Соседние файлы в папке Библиотека им академика М.И. Перельмана
