A joint ESTRO and AAPM guideline for development, clinical validation and reporting of artificial intelligence models in radiation therapy.

BACKGROUND AND PURPOSE: Artificial Intelligence (AI) models in radiation therapy are being developed with increasing pace. Despite this, the radiation therapy community has not widely adopted these models in clinical practice. A cohesive guideline on how to develop, report and clinically validate AI algorithms might help bridge this gap. METHODS AND MATERIALS: A Delphi process with all co-authors was followed to determine which topics should be addressed in this comprehensive guideline. Separate sections of the guideline, including Statements, were written by subgroups of the authors and discussed with the whole group at several meetings. Statements were formulated and scored as highly recommended or recommended. RESULTS: The following topics were found most relevant: Decision making, image analysis, volume segmentation, treatment planning, patient specific quality assurance of treatment delivery, adaptive treatment, outcome prediction, training, validation and testing of AI model parameters, model availability for others to verify, model quality assurance/updates and upgrades, ethics. Key references were given together with an outlook on current hurdles and possibilities to overcome these. 19 Statements were formulated. CONCLUSION: A cohesive guideline has been written which addresses main topics regarding AI in radiation therapy. It will help to guide development, as well as transparent and consistent reporting and validation of new AI tools and facilitate adoption.

High-resolution three-dimensional dosimetry in clinically relevant volumes utilizing optically stimulated luminescence.

BACKGROUND: The continued development of new radiotherapy techniques requires dosimetry systems that satisfy increasingly rigorous requirements, such as high sensitivity, wide dose range, and high spatial resolution. An emerging requirement is the ability to read out doses in three dimensions (3D) with high precision and spatial resolution. A few dosimetry systems with 3D capabilities are available, but their application in a clinical workflow is limited for various reasons, primarily originating from their chemical nature. The search for a 3D dosimetry system with potential for clinical implementation is thus ongoing. PURPOSE: To demonstrate the capabilities of a novel optically-stimulated-luminescence (OSL)-based 3D dosimetry system capable of measuring radiation doses in clinically relevant volumes. METHODS: A laser-based readout system was used to measure dose distributions delivered by both photons and protons, utilizing the OSL from a 50 × 50 × 50 $50\times 50\times 50$  mm 3 $^3$ YSO:Ce crystal. A homogeneous treatment plan consisting of two opposing photon fields was used to establish an inhomogeneity correction map of the crystal response and demonstrated the accuracy and precision of the system. The crystal was additionally irradiated with a photon treatment plan consisting of three overlapping 10 × 10 $10\times 10$  mm 2 $^2$ fields delivered from different angles, and a proton treatment plan consisting of four pencil beams with energies 90 MeV ( × 2 $\times 2$ ), 115 MeV, and 140 MeV. The system abilities were quantified by comparing the 3D-resolved measurements to Monte Carlo simulations. RESULTS: The dose map reproducibility of the system was found to be within 2% including both statistical and systematic errors. The measurements yielded integrated doses from a volume of 50 × 50 × 40 $50\times 50\times 40$  mm 3 $^3$ with voxel volumes of just 0.28 × 0.28 × 0.50 $0.28\times 0.28\times 0.50$  mm 3 $^3$ . An excellent agreement between the 3D-resolved measurements and the simulations was found for both photon- and proton-irradiation. CONCLUSIONS: The capabilities of the devised system for measuring clinically relevant fields of photons and proton pencil beams within a clinically relevant volume were demonstrated. The system poses as a promising candidate for clinical applications, and enables future research in the field of OSL-based tissue-equivalent 3D dosimetry.

Risk of second primary cancer from proton arc therapy of pediatric brain tumors.

Proton arc therapy (PAT) is currently explored for clinical implementation, despite its associated low-dose bath. This study therefore aimed at evaluating the risk of radiation-induced second primary cancer (SPC) for PAT in pediatric brain tumor patients. Two brain-specific models for SPC induction were applied in five cases to compare volumetric modulated arc therapy (VMAT), intensity modulated proton therapy (IMPT) and PAT surrogate plans. The PAT integral dose was reduced by a median of 29% compared to VMAT, and 17% compared to IMPT. For both models, the estimated SPC risks were consistently the lowest for PAT.

Linear energy transfer-inclusive models of brainstem necrosis following proton therapy of paediatric ependymoma.

Background and Purpose: Radiation-induced brainstem necrosis after proton therapy is a severe toxicity with potential association to uncertainties in the proton relative biological effectiveness (RBE). A constant RBE of 1.1 is assumed clinically, but the RBE is known to vary with linear energy transfer (LET). LET-inclusive predictive models of toxicity may therefore be beneficial during proton treatment planning. Hence, we aimed to construct models describing the association between brainstem necrosis and LET in the brainstem. Materials and methods: A matched case-control cohort (n = 28, 1:3 case-control ratio) of symptomatic brainstem necrosis was selected from 954 paediatric ependymoma brain tumour patients treated with passively scattered proton therapy. Dose-averaged LET (LETd) parameters in restricted volumes (L50%, L10% and L0.1cm3, the cumulative LETd) within high-dose thresholds were included in linear- and logistic regression normal tissue complication probability (NTCP) models. Results: A 1 keV/µm increase in L10% to the brainstem volume receiving dose over 54 Gy(RBE) led to an increased brainstem necrosis risk [95% confidence interval] of 2.5 [0.0, 7.8] percentage points. The corresponding logistic regression model had area under the receiver operating characteristic curve (AUC) of 0.76, increasing to 0.84 with the anterior pons substructure as a second parameter. 19 [7, 350] patients with toxicity were required to associate the L10% (D > 54 Gy(RBE)) and brainstem necrosis with 80% statistical power. Conclusion: The established models of brainstem necrosis illustrate a potential impact of high LET regions in patients receiving high doses to the brainstem, and thereby support LET mitigation during clinical treatment planning.

Dose perturbations in proton pencil beam delivery investigated by dynamically deforming silicone-based radiochromic dosimeters.

Objective.Proton therapy with pencil beam delivery enables dose distributions that conform tightly to the shape of a target. However, proton therapy dose delivery is sensitive to motion and deformation, which especially occur in the abdominal and thoracic regions. In this study, the dose perturbation caused by dynamic motion with and without gating during proton pencil beam deliveries were investigated using deformable three-dimensional (3D) silicone-based radiochromic dosimeters.Approach.A spread-out Bragg peak formed by four proton spots with different energies was delivered to two dosimeter batches. All dosimeters were cylindrical with a 50 mm diameter and length. The dosimeters were irradiated stationary while uncompressed and during dynamic compression by sinusoidal motion with peak-to-peak amplitudes of 20 mm in one end of the dosimeter and 10 mm in the other end. Motion experiments were made without gating and with gating near the uncompressed position. The entire experiment was video recorded and simulated in a Monte Carlo (MC) program.Main results.The 2%/2 mm gamma index analysis between the dose measurements and the MC dose simulations had pass rates of 86%-94% (first batch) and 98%-99% (second batch). Compared to the static delivery, the dose delivered during motion had gamma pass rates of 99%-100% when employing gating and 68%-87% without gating in the experiments whereas for the MC simulations it was 100% with gating and 66%-82% without gating.Significance.This study demonstrated the ability of using deformable 3D dosimeters to measure dose perturbations in proton pencil beam deliveries caused by dynamic motion and deformation.

A case-control study of linear energy transfer and relative biological effectiveness related to symptomatic brainstem toxicity following pediatric proton therapy.

BACKGROUND AND PURPOSE: A fixed relative biological effectiveness (RBE) of 1.1 (RBE1.1) is used clinically in proton therapy even though the RBE varies with properties such as dose level and linear energy transfer (LET). We therefore investigated if symptomatic brainstem toxicity in pediatric brain tumor patients treated with proton therapy could be associated with a variable LET and RBE. MATERIALS AND METHODS: 36 patients treated with passive scattering proton therapy were selected for a case-control study from a cohort of 954 pediatric brain tumor patients. Nine children with symptomatic brainstem toxicity were each matched to three controls based on age, diagnosis, adjuvant therapy, and brainstem RBE1.1 dose characteristics. Differences across cases and controls related to the dose-averaged LET (LETd) and variable RBE-weighted dose from two RBE models were analyzed in the high-dose region. RESULTS: LETd metrics were marginally higher for cases vs. controls for the majority of dose levels and brainstem substructures. Considering areas with doses above 54 Gy(RBE1.1), we found a moderate trend of 13% higher median LETd in the brainstem for cases compared to controls (P =.08), while the difference in the median variable RBE-weighted dose for the same structure was only 2% (P =.6). CONCLUSION: Trends towards higher LETd for cases compared to controls were noticeable across structures and LETd metrics for this patient cohort. While case-control differences were minor, an association with the observed symptomatic brainstem toxicity cannot be ruled out.

Optically stimulated luminescence in state-of-the-art LYSO:Ce scintillators enables high spatial resolution 3D dose imaging.

In this contribution, we study the optically stimulated luminescence (OSL) exhibited by commercial [Formula: see text]:Ce crystals. This photon emission mechanism, complementary to scintillation, can trap a fraction of radiation energy deposited in the material and provides sufficient signal to develop a novel post-irradiation 3D dose readout. We characterize the OSL emission through spectrally and temporally resolved measurements and monitor the dose linearity response over a broad range. The measurements show that the [Formula: see text] centers responsible for scintillation also function as recombination centers for the OSL mechanism. The capture to OSL-active traps competes with scintillation originating from the direct non-radiative energy transfer to the luminescent centers. An OSL response on the order of 100 ph/MeV is estimated. We demonstrate the imaging capabilities provided by such an OSL photon yield using a proof-of-concept optical readout method. A 0.1 [Formula: see text] spatial resolution for doses as low as 0.5 Gy is projected using a cubic crystal to image volumetric dose profiles. While OSL degrades the intrinsic scintillating performance by reducing the number of scintillation photons emitted following the passage of ionizing radiation, it can encode highly resolved spatial information of the interaction point of the particle. This feature combines ionizing radiation spectroscopy and 3D reusable dose imaging in a single material.

A Novel Nanocomposite Material for Optically Stimulated Luminescence Dosimetry.

Radiotherapy is a well-established and important treatment for cancer tumors, and advanced technologies can deliver doses in complex three-dimensional geometries tailored to each patient's specific anatomy. A 3D dosimeter, based on optically stimulated luminescence (OSL), could provide a high accuracy and reusable tool for verifying such dose delivery. Nanoparticles of an OSL material embedded in a transparent matrix have previously been proposed as an inexpensive dosimeter, which can be read out using laser-based methods. Here, we show that Cu-doped LiF nanocubes (nano-LiF:Cu) are excellent candidates for 3D OSL dosimetry owing to their high sensitivity, dose linearity, and stability at ambient conditions. We demonstrate a scalable synthesis technique producing a material with the attractive properties of a single dosimetric trap and a single near-ultraviolet emission line well separated from visible-light stimulation sources. The observed transparency and light yield of silicone sheets with embedded nanocubes hold promise for future 3D OSL-based dosimetry.

Towards range-guidance in proton therapy to detect organ motion-induced dose degradations.

Introduction.Internal organ motion and deformations may cause dose degradations in proton therapy (PT) that are challenging to resolve using conventional image-guidance strategies. This study aimed to investigate the potential ofrange guidanceusing water-equivalent path length (WEPL) calculations to detect dose degradations occurring in PT.Materials and methods. Proton ranges were estimated using WEPL calculations. Field-specific isodose surfaces in the planning CT (pCT), from robustly optimised five-field proton plans (opposing lateral and three posterior/posterior oblique beams) for locally advanced prostate cancer patients, were used as starting points. WEPLs to each point on the field-specific isodoses in the pCT were calculated. The corresponding range for each point was found in the repeat CTs (rCTs). The spatial agreement between the resulting surfaces in the rCTs (hereafter referred to as iso-WEPLs) and the isodoses re-calculated in rCTs was evaluated for different dose levels and Hausdorff thresholds (2-5 mm). Finally, the sensitivity and specificity of detecting target dose degradation (V95% < 95%) using spatial agreement measures between the iso-WEPLs and isodoses in the pCT was evaluated.Results. The spatial agreement between the iso-WEPLs and isodoses in the rCTs depended on the Hausdorff threshold. The agreement was 65%-88% for a 2 mm threshold, 83%-96% for 3 mm, 90%-99% for 4 mm, and 94%-99% for 5 mm, across all fields and isodose levels. Minor differences were observed between the different isodose levels investigated. Target dose degradations were detected with 82%-100% sensitivity and 75%-80% specificity using a 2 mm Hausdorff threshold for the lateral fields.Conclusion. Iso-WEPLs were comparable to isodoses re-calculated in the rCTs. The proposed strategy could detect target dose degradations occurring in the rCTs and could be an alternative to a fully-fledged dose re-calculation to detect anatomical variations severely influencing the proton range.

A phase I/II study of acute and late physician assessed and patient-reported morbidity following whole pelvic radiation in high-risk prostate cancer patients.

BACKGROUND: The aim of this study was to assess acute and late morbidity measured by the physician and patient-reported outcomes (PROs) in high-risk prostate cancer (PC) patients receiving whole pelvic intensity-modulated radiotherapy (IMRT) in the setting of a national clinical trial. MATERIAL AND METHODS: A total of 88 patients with adenocarcinoma of the prostate and high-risk parameters were enrolled from 2011 to 2013. All patients received 78 Gy in 39 fractions of IMRT delivering simultaneous 78 Gy to the prostate and 56 Gy to the seminal vesicles and lymph nodes. Physician-reported morbidity was assessed by CTCAE v.4.0. PROs were registered for gastro-intestinal (GI) by the RT-ARD score, genito-urinary (GU) by DAN-PSS, sexual and hormonal by EPIC-26, and quality of life (QoL) by EORTC QLQ-C30. RESULTS: Median follow-up (FU) time was 4.6 years. No persistent late CTCAE grade 3+ morbidity was observed. Prevalence of CTCAE grade 2+ GI morbidities varied from 0 to 6% at baseline throughout FU time, except for diarrhea, which was reported in 19% of the patients post-RT. PROs revealed increased GI morbidity (≥1 monthly episode) for "rectal urgency", "use of pads", "incomplete evacuation", "mucus in stool" and "bowel function impact on QoL" all remained significantly different (p < .05) at 60 months compared to baseline. CTCAE grade 2+ GU and sexual morbidity were unchanged. GU PROs on obstructive and irritative GU items (≥daily episode) increased during RT and normalized at 24 months. No clinically significant differences were found in sexual, hormonal, and QoL scores compared to baseline. CONCLUSIONS: Whole pelvic RT resulted in a mild to the moderate burden of late GI morbidities demonstrated by a relatively high prevalence of PROs. Whereas, physician-assessed morbidity revealed a low prevalence of late GI morbidity scores. This emphasizes the importance of using both PROs and physician-reported scoring scales when reporting late morbidity in clinical trials.

Normal tissue complication probability models for prospectively scored late rectal and urinary morbidity after proton therapy of prostate cancer.

BACKGROUND AND PURPOSE: Photons and protons have fundamentally different properties, i.e. protons have a reduced dose bath but a higher relative biological effectiveness. Photon-based normal tissue complication probability (NTCP) models may therefore not immediately be applicable to proton therapy (PT). The aim was to derive parameters of the Lyman-Kutcher-Burman (LKB) NTCP model using prospectively recorded late morbidity data from PT, focusing on rectal morbidity and prostate cancer. MATERIALS AND METHODS: Prospectively collected data were available for 1151 prostate cancer patients treated with passive scattering PT and prescribed target doses of 78-82 Gy (RBE = 1.1) in 2 Gy fractions. Morbidity data (CTCAE v3.0) consisted of two alternative late grade 2 rectal bleeding endpoints: Medical Grade2A (GR2A) and procedural Grade2B (GR2B), as well as late grade 3 + urinary morbidity. GR2A + 2B were observed in 156/1047 patients (15%), GR2B in 45/1047 patients (4%), and urinary grade 3 + in 51/1151 patients (4%). LKB NTCP model parameters (D50, m, and n) were derived by maximum likelihood estimation. RESULTS: For the rectum/rectal wall the volume parameter n was low (0.07-0.14) for both GR2A + 2B and GR2B, as was the m parameter (range: 0.16-0.20). For the bladder/bladder wall both parameters were high (n-range: 0.20-0.36; m-range: 0.32-0.36). D50 parameters were higher for GR2B of the rectum/rectal wall (95.9-98.0 Gy) and bladder/bladder wall (118.1-119.9 Gy), but lower for GR2A2B (71.7-73.6 Gy). CONCLUSION: PT specific LKB NTCP model parameters were derived from a population of more than 1000 patients. The D50 parameter differed for all structures and endpoints and deviated from typical photon-based LKB model values.