Patient selection for proton therapy using Normal Tissue Complication Probability with deep learning dose prediction for oropharyngeal cancer.

BACKGROUND: In cancer care, determining the most beneficial treatment technique is a key decision affecting the patient's survival and quality of life. Patient selection for proton therapy (PT) over conventional radiotherapy (XT) currently entails comparing manually generated treatment plans, which requires time and expertise. PURPOSE: We developed an automatic and fast tool, AI-PROTIPP (Artificial Intelligence Predictive Radiation Oncology Treatment Indication to Photons/Protons), that assesses quantitatively the benefits of each therapeutic option. Our method uses deep learning (DL) models to directly predict the dose distributions for a given patient for both XT and PT. By using models that estimate the Normal Tissue Complication Probability (NTCP), namely the likelihood of side effects to occur for a specific patient, AI-PROTIPP can propose a treatment selection quickly and automatically. METHODS: A database of 60 patients presenting oropharyngeal cancer, obtained from the Cliniques Universitaires Saint Luc in Belgium, was used in this study. For every patient, a PT plan and an XT plan were generated. The dose distributions were used to train the two dose DL prediction models (one for each modality). The model is based on U-Net architecture, a type of convolutional neural network currently considered as the state of the art for dose prediction models. A NTCP protocol used in the Dutch model-based approach, including grades II and III xerostomia and grades II and III dysphagia, was later applied in order to perform automatic treatment selection for each patient. The networks were trained using a nested cross-validation approach with 11-folds. We set aside three patients in an outer set and each fold consists of 47 patients in training, five in validation and five for testing. This method allowed us to assess our method on 55 patients (five patients per test times the number of folds). RESULTS: The treatment selection based on the DL-predicted doses reached an accuracy of 87.4% for the threshold parameters set by the Health Council of the Netherlands. The selected treatment is directly linked with these threshold parameters as they express the minimal gain brought by the PT treatment for a patient to be indicated to PT. To validate the performance of AI-PROTIPP in other conditions, we modulated these thresholds, and the accuracy was above 81% for all the considered cases. The difference in average cumulative NTCP per patient of predicted and clinical dose distributions is very similar (less than 1% difference). CONCLUSIONS: AI-PROTIPP shows that using DL dose prediction in combination with NTCP models to select PT for patients is feasible and can help to save time by avoiding the generation of treatment plans only used for the comparison. Moreover, DL models are transferable, allowing, in the future, experience to be shared with centers that would not have PT planning expertise.

Artificial intelligence and machine learning for medical imaging: A technology review.

Artificial intelligence (AI) has recently become a very popular buzzword, as a consequence of disruptive technical advances and impressive experimental results, notably in the field of image analysis and processing. In medicine, specialties where images are central, like radiology, pathology or oncology, have seized the opportunity and considerable efforts in research and development have been deployed to transfer the potential of AI to clinical applications. With AI becoming a more mainstream tool for typical medical imaging analysis tasks, such as diagnosis, segmentation, or classification, the key for a safe and efficient use of clinical AI applications relies, in part, on informed practitioners. The aim of this review is to present the basic technological pillars of AI, together with the state-of-the-art machine learning methods and their application to medical imaging. In addition, we discuss the new trends and future research directions. This will help the reader to understand how AI methods are now becoming an ubiquitous tool in any medical image analysis workflow and pave the way for the clinical implementation of AI-based solutions.

Deep learning dose prediction for IMRT of esophageal cancer: The effect of data quality and quantity on model performance.

PURPOSE: To investigate the effect of data quality and quantity on the performance of deep learning (DL) models, for dose prediction of intensity-modulated radiotherapy (IMRT) of esophageal cancer. MATERIAL AND METHODS: Two databases were used: a variable database (VarDB) with 56 clinical cases extracted retrospectively, including user-dependent variability in delineation and planning, different machines and beam configurations; and a homogenized database (HomDB), created to reduce this variability by re-contouring and re-planning all patients with a fixed class-solution protocol. Experiment 1 analysed the user-dependent variability, using 26 patients planned with the same machine and beam setup (E26-VarDB versus E26-HomDB). Experiment 2 increased the training set by groups of 10 patients (E16, E26, E36, E46, and E56) for both databases. Model evaluation metrics were the mean absolute error (MAE) for selected dose-volume metrics and the global MAE for all body voxels. RESULTS: For Experiment 1, E26-HomDB reduced the MAE for the considered dose-volume metrics compared to E26-VarDB (e.g. reduction of 0.2 Gy for D95-PTV, 1.2 Gy for Dmean-heart or 3.3% for V5-lungs). For Experiment 2, increasing the database size slightly improved performance for HomDB models (e.g. decrease in global MAE of 0.13 Gy for E56-HomDB versus E26-HomDB), but increased the error for the VarDB models (e.g. increase in global MAE of 0.20 Gy for E56-VarDB versus E26-VarDB). CONCLUSION: A small database may suffice to obtain good DL prediction performance, provided that homogenous training data is used. Data variability reduces the performance of DL models, which is further pronounced when increasing the training set.

Improvement of kilovoltage intrafraction monitoring accuracy through gantry angles selection.

Kilovoltage intrafraction monitoring (KIM) is a method allowing to precisely infer the tumour trajectory based on cone beam computed tomography (CBCT) 2D-projections. However, its accuracy is deteriorated in the case of highly mobile tumours involving hysteresis. A first adaptation of KIM consisting of a prior amplitude based binning step has been developed in order to minimize the errors of the original model (phase-KIM). In this work, we propose enhanced methods (KIMsub-arc optimand phase-KIMsub-arc optim) to improve the accuracy of KIM and phase-KIM which relies on the selection of the optimal starting CBCT gantry angle. Aiming at demonstrating the interest of our approach, we carried out a simulation study and an experimental study: we compared the accuracy of the conventional versus sub-arc optim methods on simulated realistic tumour motions with amplitudes ranging from 5 to 30 mm in 1 mm increments. The same approach was performed using a lung dynamic phantom generating a 30 mm amplitude sinusoidal motion. The results show that for in-silico simulated motions of 10, 20 and 30 mm amplitude, the three-dimensional root mean square error (3D-RMSE) can be reduced by 0.67 mm, 0.91 mm, 0.94 mm and 0.18 mm, 0.25 mm, 0.28 mm using KIMsub-arc optimand phase-KIMsub-arc optimrespectively. Considering all in-silico simulated trajectories, the percentage of errors larger than 1 mm decreases from 21.9% down to 1.6% for KIM (p < 0.001) and from 6.6% down to 1.2% for phase-KIM (p < 0.001). Experimentally, the 3D-RMSE is lowered by 0.5732 mm for KIM and by 0.1 mm for phase-KIM. The percentage of errors larger than 1 mm falls from 39.7% down to 18.5% for KIM and from 23.2% down to 11.1% for phase-KIM. In conclusion, our method efficiently anticipates CBCT gantry angles associated with a significantly better accuracy by using KIM and phase-KIM.

Validation and IMRT/VMAT delivery quality of a preconfigured fast-rotating O-ring linac system.

PURPOSE: A fast-rotating O-ring dedicated intensity modulated radiotherapy (IMRT)/volumetric modulated arc therapy (VMAT) delivery system, the Halcyon, is delivered by default with a fully preconfigured photon beam model in the treatment planning system (TPS). This work reports on the validation and achieved IMRT/VMAT delivery quality on the system. METHODS: Acceptance testing followed the vendor's installation product acceptance and was supplemented with mechanical QA. The dosimetric calibration was performed according to the IAEA TRS-398 code-of-practice, delivering 600 cGy/min at 10 cm depth, a 90 cm source-surface distance, and a 10 × 10 cm² field size. The output factors, multileaf collimator (MLC) transmission and dosimetric leaf gap (DLG) were validated by comparing measurements with the modeled values in the TPS. Validation of IMRT/VMAT was conducted following AAPM reports (MPPG 5.a, TG-119). Next, dose measurements were performed for end-to-end (E2E) checks in heterogeneous anthropomorphic phantoms using radiochromic film in multiple planes and using ionization chambers (IC) point measurements. E2E checks were performed for VMAT (cranial, rectum, spine, and head and neck) and IMRT (lung). Additionally, IROC Houston mailed dosimetry audits were performed for the beam calibration and E2E measurements using a thorax phantom (IMRT) and a head and neck phantom (VMAT). Lastly, extensive patient-specific QA was performed for the first patients of each new indication, 26 in total (nrectum = 2, nspine = 5, nlung = 5, nesophagus = 2, nhead and neck = 7, ncranial = 5), treated on the fast-rotating O-ring linac. The patient-specific QA followed the AAPM TG-218 guidelines and comprised of portal dosimetry, ArcCHECK diode array, radiochromic film dosimetry in a MultiCube phantom, and IC point measurements. RESULTS: The measured output factors showed an agreement <1% for fields ≥3 × 3 cm². Field sizes ≤2 × 2 cm² had a difference of <2%. The measured single-layer MLC transmission was 0.42 ± 0.01% and the measured DLG was 0.27 ± 0.22 mm. The AAPM MPPG 5.a measurements were fully compliant with the guideline criteria. Dose differences larger than 2% were found for the PDD at large depths (>25 cm). TG-119's confidence limits were achieved for the VMAT point dose measurements and for both the IMRT and VMAT radiochromic film measurements. The TG-119 confidence limits were not achieved for IMRT point dose measurements in both the target (5.9%) and the avoidance structure (6.4%). All E2E tests had point differences below 2.3% and gamma agreement scores above 90.6%. The IROC beam calibration audit showed agreement of <1%. The IROC lung IMRT audit and head and neck VMAT audit had results compliant with the IROC Houston's credentialing criteria. All IMRT and VMAT plans selected for patient-specific QA were within the action limits suggested by TG-218. CONCLUSIONS: The fast-rotating O-ring linac and its preconfigured TPS are compliant with the international commissioning criteria of AAPM MPPG 5.a and AAPM TG-119. E2E measurements on heterogeneous anthropomorphic phantoms were within clinically acceptable tolerances. IROC Houston's audits satisfied the credentialing criteria. This work comprises the first extensive dataset reporting on the preconfigured fast-rotating O-ring linac.

Development and accuracy evaluation of a single-camera intra-bore surface scanning system for radiotherapy in an O-ring linac.

BACKGROUND AND PURPOSE: Current commercial surface scanning systems are not able to monitor patients during radiotherapy fractions in closed-bore linacs during adaptive workflows. In this work a surface scanning system for monitoring in an O-ring linac is proposed. METHODS AND MATERIALS: A depth camera was mounted at the backend of the bore. The acquired surface point cloud was transformed to the linac coordinate system after a cube detection calibration step. The real-time surface was registered using an Iterative Closest Point algorithm to a reference region-of-interest of the body contour from the planning CT and of a depth camera surface acquisition from the first fraction. The positioning accuracy was investigated using anthropomorphic 3D-printed phantoms with embedded markers: a head, hand and breast. To simulate clinically observed positioning errors, each phantom was placed 24 times with 0-10 mm and 0-8° offsets from the planned position. At every position a cone-beam CT (CBCT) was acquired and a surface registration performed. The surface registration error was determined as the difference between the surface registration and the CBCT-to-CT fiducial marker registration. RESULTS: The registration errors were (mean ±â€¯SD): lat: 0.4 ±â€¯0.8 mm, vert: -0.2 ±â€¯0.2 mm, long: 0.3 ±â€¯0.5 mm and Yaw: -0.2 ±â€¯0.6°, Pitch: 0.4 ±â€¯0.2°, Roll: 0.5 ±â€¯0.8° for the body contour reference, and lat: -0.7 ±â€¯0.7 mm, vert: 0.3 ±â€¯0.2 mm, long: 0.2 ±â€¯0.5 mm and Yaw: -0.5 ±â€¯0.5°, Pitch: 0.1 ±â€¯0.3°, Roll: -0.7 ±â€¯0.7° for the captured surface reference. CONCLUSION: The proposed single camera intra-bore surface system was capable of accurately detecting phantom displacements and allows intrafraction motion monitoring for surface guided radiotherapy inside the bore of O-ring gantries.

Production of patient-specific electron beam aperture cut-outs using a low-cost, multi-purpose 3D printer.

Electron beam collimators for non-standard field sizes and shapes are typically fabricated using Styrofoam molds to cast the aperture cut-out. These molds are often produced using a dedicated foam cutter, which may be expensive and only serves a single purpose. An increasing number of radiotherapy departments, however, has a 3D printer on-site, to create a wide range of custom-made treatment auxiliaries, such as bolus and dosimetry phantoms. The 3D printer can also be used to produce patient-specific aperture cut-outs, as elaborated in this note. Open-source programming language was used to automatically generate the mold's shape in a generic digital file format readable by 3D printer software. The geometric mold model has the patient's identification number integrated and is to be mounted on a uniquely fitting, reusable positioning device, which can be 3D printed as well. This assembly likewise fits uniquely onto the applicator tray, ensuring correct and error-free alignment of the mold during casting of the aperture. For dosimetric verification, two aperture cut-outs were cast, one using a conventionally cut Styrofoam mold and one using a 3D printed mold. Using these cut-outs, the clinical plan was delivered onto a phantom, for which the transversal dose distributions were measured at 2 cm depth using radiochromic film and compared using gamma-index analysis. An agreement score of 99.9% between the measured 2D dose distributions was found in the (10%-80%) dose region, using 1% (local) dose-difference and 1.0 mm distance-to-agreement acceptance criteria. The workflow using 3D printing has been clinically implemented and is in routine use at the author's institute for all patient-specific electron beam aperture cut-outs. It allows for a standardized, cost-effective, and operator-friendly workflow without the need for dedicated equipment. In addition, it offers possibilities to increase safety and quality of the process including patient identification and methods for accurate mold alignment.

Volumetric modulated arc therapy of head-and-neck cancer on a fast-rotating O-ring linac: Plan quality and delivery time comparison with a C-arm linac.

BACKGROUND AND PURPOSE: Linac improvements in gantry speed, leaf speed and dose rate may increase the time-efficiency of volumetric modulated arc therapy (VMAT) delivery. The plan quality achievable with faster VMAT however remains to be investigated. In this study, a fast-rotating O-ring linac with fast-moving leaves is compared with a C-arm linac in terms of plan quality and delivery time for VMAT of head-and-neck cancer (HNC). MATERIAL AND METHODS: For 30 patients with HNC, treatment planning was performed using dual-arc (HA2) and triple-arc (HA3) VMAT on a Halcyon fast-rotating O-ring linac and using dual-arc VMAT on a TrueBeam C-arm linac (TB2). Target coverage metrics and complication probabilities were compared. Plan delivery was verified using 3%/3 mm gamma-index analysis of helical diode array measurements. Volumetric image acquisition and plan delivery times were compared. RESULTS: All studied VMAT-techniques fulfilled the target coverage objectives. D2% to the boost volume was higher for HA2 (median 103.7%, 1st-3rd quartile [103.5%;104.0%]) and HA3 (103.2% [103.0%;103.7%)] than for TB2 (102.6% [102.3%;103.0%)], resulting in an increased boost target dose heterogeneity for HA2 and HA3. Complication probabilities were comparable between HA2 and TB2, while HA3 showed a xerostomia probability reduction (0.8% [0.2%;1.8%]) and dysphagia probability reduction (1.0% [0.2%;1.8%]) compared with TB2. Gamma-index agreement scores were never below 93.0% for HA2, HA3 and TB2. Volumetric imaging and plan delivery time was shorter for HA2 (1 m 24 s ±â€¯1 s) and HA3 (1 m 54 s ±â€¯1 s) than for TB2 (2 m 47 s ±â€¯1 s). CONCLUSION: For VMAT of HNC, the fast-rotating O-ring linac at least maintains the plan quality of two arcs on a C-arm linac while reducing the image acquisition and plan delivery time.

Patient-specific bolus for range shifter air gap reduction in intensity-modulated proton therapy of head-and-neck cancer studied with Monte Carlo based plan optimization.

BACKGROUND & PURPOSE: Intensity-modulated proton therapy (IMPT) of superficial lesions requires pre-absorbing range shifter (RS) to deliver the more shallow spots. RS air gap minimization is important to avoid spot size degradation, but remains challenging in complex geometries such as in head-and-neck cancer (HNC). In this study, clinical endpoints were investigated for patient-specific bolus and for conventional RS solutions, making use of a Monte Carlo (MC) dose engine for IMPT optimization. METHODS AND MATERIALS: For 5 oropharyngeal cancer patients, IMPT spot maps were generated using beamlets calculated with MC. The plans were optimized for three different RS configurations: 3D printed on-skin bolus, snout- and nozzle-mounted RS. Organ-at-risk (OAR) doses and late toxicity probabilities were compared between all configuration-specific optimized plans. RESULTS: The use of bolus reduced the mean dose to all OARs compared to snout and nozzle-mounted RS. The contralateral parotid gland and supraglottic larynx received on average 2.9Gy and 4.2Gy less dose compared to the snout RS. Bolus reduced the average probability for xerostomia by 3.0%. For dysphagia, bolus reduced the probability by 2.7%. CONCLUSIONS: Quantification of the dosimetric advantage of patient-specific bolus shows significant reductions compared to conventional RS solutions for xerostomia and dysphagia probability. These results motivate the development of a patient-specific bolus solution in IMPT for HNC.

Towards 3D printed multifunctional immobilization for proton therapy: Initial materials characterization.

PURPOSE: 3D printing technology is investigated for the purpose of patient immobilization during proton therapy. It potentially enables a merge of patient immobilization, bolus range shifting, and other functions into one single patient-specific structure. In this first step, a set of 3D printed materials is characterized in detail, in terms of structural and radiological properties, elemental composition, directional dependence, and structural changes induced by radiation damage. These data will serve as inputs for the design of 3D printed immobilization structure prototypes. METHODS: Using four different 3D printing techniques, in total eight materials were subjected to testing. Samples with a nominal dimension of 20 × 20 × 80 mm3 were 3D printed. The geometrical printing accuracy of each test sample was measured with a dial gage. To assess the mechanical response of the samples, standardized compression tests were performed to determine the Young's modulus. To investigate the effect of radiation on the mechanical response, the mechanical tests were performed both prior and after the administration of clinically relevant dose levels (70 Gy), multiplied with a safety factor of 1.4. Dual energy computed tomography (DECT) methods were used to calculate the relative electron density to water ρe, the effective atomic number Zeff, and the proton stopping power ratio (SPR) to water SPR. In order to validate the DECT based calculation of radiological properties, beam measurements were performed on the 3D printed samples as well. Photon irradiations were performed to measure the photon linear attenuation coefficients, while proton irradiations were performed to measure the proton range shift of the samples. The directional dependence of these properties was investigated by performing the irradiations for different orientations of the samples. RESULTS: The printed test objects showed reduced geometric printing accuracy for 2 materials (deviation > 0.25 mm). Compression tests yielded Young's moduli ranging from 0.6 to 2940 MPa. No deterioration in the mechanical response was observed after exposure of the samples to 100 Gy in a therapeutic MV photon beam. The DECT-based characterization yielded Zeff ranging from 5.91 to 10.43. The SPR and ρe both ranged from 0.6 to 1.22. The measured photon attenuation coefficients at clinical energies scaled linearly with ρe. Good agreement was seen between the DECT estimated SPR and the measured range shift, except for the higher Zeff. As opposed to the photon attenuation, the proton range shifting appeared to be printing orientation dependent for certain materials. CONCLUSIONS: In this study, the first step toward 3D printed, multifunctional immobilization was performed, by going through a candidate clinical workflow for the first time: from the material printing to DECT characterization with a verification through beam measurements. Besides a proof of concept for beam modification, the mechanical response of printed materials was also investigated to assess their capabilities for positioning functionality. For the studied set of printing techniques and materials, a wide variety of mechanical and radiological properties can be selected from for the intended purpose. Moreover the elaborated hybrid DECT methods aid in performing in-house quality assurance of 3D printed components, as these methods enable the estimation of the radiological properties relevant for use in radiation therapy.