Novel dosimetric validation of a commercial CT scanner based deep learning automated contour solution for prostate radiotherapy.

PURPOSE: OAR delineation accuracy influences: (i) a patient's optimised dose distribution (PD), (ii) the reported doses (RD) presented at approval, which represent plan quality. This study utilised a novel dosimetric validation methodology, comprehensively evaluating a new CT-scanner-based AI contouring solution in terms of PD and RD within an automated planning workflow. METHODS: 20 prostate patients were selected to evaluate AI contouring for rectum, bladder, and proximal femurs. Five planning 'pipelines' were considered; three using AI contours with differing levels of manual editing (nominally none (AIStd), minor editing in specific regions (AIMinEd), and fully corrected (AIFullEd)). Remaining pipelines were manual delineations from two observers (MDOb1, MDOb2). Automated radiotherapy plans were generated for each pipeline. Geometric and dosimetric agreement of contour sets AIStd, AIMinEd, AIFullEd and MDOb2 were evaluated against the reference set MDOb1. Non-inferiority of AI pipelines was assessed, hypothesising that compared to MDOb1, absolute deviations in metrics for AI contouring were no greater than that from MDOb2. RESULTS: Compared to MDOb1, organ delineation time was reduced by 24.9 min (96 %), 21.4 min (79 %) and 12.2 min (45 %) for AIStd, AIMinEd and AIFullEd respectively. All pipelines exhibited generally good dosimetric agreement with MDOb1. For RD, median deviations were within ± 1.8 cm3, ± 1.7 % and ± 0.6 Gy for absolute volume, relative volume and mean dose metrics respectively. For PD, respective values were within ± 0.4 cm3, ± 0.5 % and ± 0.2 Gy. Statistically (p < 0.05), AIMinEd and AIFullEd were dosimetrically non-inferior to MDOb2. CONCLUSIONS: This novel dosimetric validation demonstrated that following targeted minor editing (AIMinEd), AI contours were dosimetrically non-inferior to manual delineations, reducing delineation time by 79 %.

Clinical Validation of Siemens' Syngo.via Automatic Contouring System.

Purpose: The manual delineation of organs at risk is a process that requires a great deal of time both for the technician and for the physician. Availability of validated software tools assisted by artificial intelligence would be of great benefit, as it would significantly improve the radiation therapy workflow, reducing the time required for segmentation. The purpose of this article is to validate the deep learning-based autocontouring solution integrated in syngo.via RT Image Suite VB40 (Siemens Healthineers, Forchheim, Germany). Methods and Materials: For this purpose, we have used our own specific qualitative classification system, RANK, to evaluate more than 600 contours corresponding to 18 different automatically delineated organs at risk. Computed tomography data sets of 95 different patients were included: 30 patients with lung, 30 patients with breast, and 35 male patients with pelvic cancer. The automatically generated structures were reviewed in the Eclipse Contouring module independently by 3 observers: an expert physician, an expert technician, and a junior physician. Results: There is a statistically significant difference between the Dice coefficient associated with RANK 4 compared with the coefficient associated with RANKs 2 and 3 (P < .001). In total, 64% of the evaluated structures received the maximum score, 4. Only 1% of the structures were classified with the lowest score, 1. The time savings for breast, thorax, and pelvis were 87.6%, 93.5%, and 82.2%, respectively. Conclusions: Siemens' syngo.via RT Image Suite offers good autocontouring results and significant time savings.

A clinical and time savings evaluation of a deep learning automatic contouring algorithm.

Automatic contouring algorithms may streamline clinical workflows by reducing normal organ-at-risk (OAR) contouring time. Here we report the first comprehensive quantitative and qualitative evaluation, along with time savings assessment for a prototype deep learning segmentation algorithm from Siemens Healthineers. The accuracy of contours generated by the prototype were evaluated quantitatively using the Sorensen-Dice coefficient (Dice), Jaccard index (JC), and Hausdorff distance (Haus). Normal pelvic and head and neck OAR contours were evaluated retrospectively comparing the automatic and manual clinical contours in 100 patient cases. Contouring performance outliers were investigated. To quantify the time savings, a certified medical dosimetrist manually contoured de novo and, separately, edited the generated OARs for 10 head and neck and 10 pelvic patients. The automatic, edited, and manually generated contours were visually evaluated and scored by a practicing radiation oncologist on a scale of 1-4, where a higher score indicated better performance. The quantitative comparison revealed high (> 0.8) Dice and JC performance for relatively large organs such as the lungs, brain, femurs, and kidneys. Smaller elongated structures that had relatively low Dice and JC values tended to have low Hausdorff distances. Poor performing outlier cases revealed common anatomical inconsistencies including overestimation of the bladder and incorrect superior-inferior truncation of the spinal cord and femur contours. In all cases, editing contours was faster than manual contouring with an average time saving of 43.4% or 11.8 minutes per patient. The physician scored 240 structures with > 95% of structures receiving a score of 3 or 4. Of the structures reviewed, only 11 structures needed major revision or to be redone entirely. Our results indicate the evaluated auto-contouring solution has the potential to reduce clinical contouring time. The algorithm's performance is promising, but human review and some editing is required prior to clinical use.

A deep image-to-image network organ segmentation algorithm for radiation treatment planning: principles and evaluation.

BACKGROUND: We describe and evaluate a deep network algorithm which automatically contours organs at risk in the thorax and pelvis on computed tomography (CT) images for radiation treatment planning. METHODS: The algorithm identifies the region of interest (ROI) automatically by detecting anatomical landmarks around the specific organs using a deep reinforcement learning technique. The segmentation is restricted to this ROI and performed by a deep image-to-image network (DI2IN) based on a convolutional encoder-decoder architecture combined with multi-level feature concatenation. The algorithm is commercially available in the medical products "syngo.via RT Image Suite VB50" and "AI-Rad Companion Organs RT VA20" (Siemens Healthineers). For evaluation, thoracic CT images of 237 patients and pelvic CT images of 102 patients were manually contoured following the Radiation Therapy Oncology Group (RTOG) guidelines and compared to the DI2IN results using metrics for volume, overlap and distance, e.g., Dice Similarity Coefficient (DSC) and Hausdorff Distance (HD95). The contours were also compared visually slice by slice. RESULTS: We observed high correlations between automatic and manual contours. The best results were obtained for the lungs (DSC 0.97, HD95 2.7 mm/2.9 mm for left/right lung), followed by heart (DSC 0.92, HD95 4.4 mm), bladder (DSC 0.88, HD95 6.7 mm) and rectum (DSC 0.79, HD95 10.8 mm). Visual inspection showed excellent agreements with some exceptions for heart and rectum. CONCLUSIONS: The DI2IN algorithm automatically generated contours for organs at risk close to those by a human expert, making the contouring step in radiation treatment planning simpler and faster. Few cases still required manual corrections, mainly for heart and rectum.

Reduction of clinical safety margins in proton therapy enabled by the clinical implementation of dual-energy CT for direct stopping-power prediction.

PURPOSE: To quantifiy the range uncertainty in proton treatment planning using dual-energy computed tomography (DECT) for a direct stopping-power prediction (DirectSPR) algorithm and its clinical implementation. METHODS AND MATERIALS: To assess the overall uncertainty in stopping-power ratio (SPR) prediction of a DirectSPR implementation calibrated for different patient geometries, the influencing factors were categorized in imaging, modeling as well as others. The respective SPR uncertainty was quantified for lung, soft tissue and bone and translated into range uncertainty for several tumor types. The amount of healthy tissue spared was quantified for 250 patients treated with DirectSPR and the dosimetric impact was evaluated exemplarily for a representative brain-tumor patient. RESULTS: For bone, soft tissue and lung, an SPR uncertainty (1σ) of 1.6%, 1.3% and 1.3% was determined for DirectSPR, respectively. This allowed for a reduction of the clinically applied range uncertainty from currently (3.5% + 2 mm) to (1.7% + 2 mm) for brain-tumor and (2.0% + 2 mm) for prostate-cancer patients. The 150 brain-tumor and 100 prostate-cancer patients treated using DirectSPR benefitted from sparing on average 2.6 mm and 4.4 mm of healthy tissue in beam direction, respectively. In the representative patient case, dose reduction in organs at risk close to the target volume was achieved, with a mean dose reduction of up to 16% in the brainstem. Patient-specific DECT-based treatment planning with reduced safety margins was successfully introduced into clinical routine. CONCLUSIONS: A substantial increase in range prediction accuracy in clinical proton treatment planning was achieved by patient-specific DECT-based SPR prediction. For the first time, a relevant imaging-based reduction of range prediction uncertainty on a 2% level has been achieved.

Refinement of the Hounsfield look-up table by retrospective application of patient-specific direct proton stopping-power prediction from dual-energy CT.

BACKGROUND AND PURPOSE: Proton treatment planning relies on an accurate determination of stopping-power ratio (SPR) from x-ray computed tomography (CT). A refinement of the heuristic CT-based SPR prediction using a state-of-the-art Hounsfield look-up table (HLUT) is proposed, which incorporates patient SPR information obtained from dual-energy CT (DECT) in a retrospective patient-cohort analysis. MATERIAL AND METHODS: SPR datasets of 25 brain-tumor patients, 25 prostate-cancer patients, and three nonsmall cell lung-cancer (NSCLC) patients were calculated from clinical DECT scans with the comprehensively validated DirectSPR approach. Based on the median frequency distribution of voxelwise correlations between CT number and SPR within the irradiated volume, a piecewise linear function was specified (DirectSPR-based adapted HLUT). Differences in dose distribution and proton range were assessed for the nonadapted and adapted HLUT in comparison to the DirectSPR method, which has been shown to be an accurate and reliable SPR estimation method. RESULTS: The application of the DirectSPR-based adapted HLUT instead of the nonadapted HLUT reduced the systematic proton range differences from 1.2% (1.1 mm) to -0.1% (0.0 mm) for brain-tumor patients, 1.7% (4.1 mm) to 0.2% (0.5 mm) for prostate-cancer patients, and 2.0% (2.9 mm) to -0.1% (0.0 mm) for NSCLC patients. Due to the large intra- and inter-patient tissue variability, range differences to DirectSPR larger than 1% remained for the adapted HLUT. CONCLUSIONS: The incorporation of patient-specific correlations between CT number and SPR, derived from a retrospective application of DirectSPR to a broad patient cohort, improves the SPR accuracy of the current state-of-the-art HLUT approach. The DirectSPR-based adapted HLUT has been clinically implemented at the University Proton Therapy Dresden (Dresden, Germany) in 2017. This already facilitates the benefits of an improved DECT-based tissue differentiation within clinical routine without changing the general approach for range prediction (HLUT), and represents a further step toward full integration of the DECT-based DirectSPR method for treatment planning in proton therapy.

Towards a generalised development of synthetic CT images and assessment of their dosimetric accuracy.

Background: The interest in generating "synthetic computed tomography (CT) images" from magnetic resonance (MR) images has been increasing over the past years due to advances in MR guidance for radiotherapy. A variety of methods for synthetic CT creation have been developed, from simple bulk density assignment to complex machine learning algorithms.Material and methods: In this study, we present a general method to determine simplistic synthetic CTs and evaluate them according to their dosimetric accuracy. It separates the requirements on the MR image and the associated calculation effort to generate a synthetic CT. To evaluate the significance of the dosimetric accuracy under realistic conditions, clinically common uncertainties including position shifts and Hounsfield lookup table (HLUT) errors were simulated. To illustrate our approach, we first translated CT images from a test set of six pelvic cancer patients to relative electron density (ED) via a clinical HLUT. For each patient, seven simplified ED images (simED) were generated at different levels of complexity, ranging from one to four tissue classes. Then, dose distributions optimised on the reference ED image and the simEDs were compared to each other in terms of gamma pass rates (2 mm/2% criteria) and dose volume metrics.Results: For our test set, best results were obtained for simEDs with four tissue classes representing fat, soft tissue, air, and bone. For this simED, gamma pass rates of 99.95% (range: 99.72-100%) were achieved. The decrease in accuracy from ED simplification was smaller in this case than the influence of the uncertainty scenarios on the reference image, both for gamma pass rates and dose volume metrics.Conclusions: The presented workflow helps to determine the required complexity of synthetic CTs with respect to their dosimetric accuracy. The investigated cases showed potential simplifications, based on which the synthetic CT generation could be faster and more reproducible.

Dual-Energy Computed Tomography to Assess Intra- and Inter-Patient Tissue Variability for Proton Treatment Planning of Patients With Brain Tumor.

PURPOSE: Range prediction in particle therapy is associated with an uncertainty originating from calculating the stopping-power ratio (SPR) based on x-ray computed tomography (CT). Here, we assessed the intra- and inter-patient variability of tissue properties in patients with primary brain tumor using dual-energy CT (DECT) and quantified its influence on current SPR prediction. METHODS AND MATERIALS: For 102 patients' DECT scans, SPR distributions were derived from a patient-specific DECT-based approach (DirectSPR). The impact of soft tissue diversity and age-related variations in bone composition on SPR were assessed. Tissue-specific and global deviations between this method and the state-of-the-art CT-number-to-SPR conversion applying a Hounsfield look-up table (HLUT) were quantified. To isolate systematic deviations between the two, the HLUT was also optimized using DECT information. RESULTS: An intra-patient ± inter-patient soft tissue diversity of 5.6% ± 0.7% in SPR (width of 95% confidence interval) was obtained including imaging- and model-related variations of up to 2.9%. This intra-patient SPR variability is associated with a mean absolute SPR deviation of 1.2% between the patient-specific DirectSPR approach and an optimal HLUT. Between adults and children younger than 6 years, age-related variations in bone composition resulted in a median SPR difference of approximately 5%. CONCLUSIONS: Accurate patient-specific DECT-based stopping-power prediction allows for improved handling of tissue mixtures and can intrinsically incorporate most of the SPR variability arising from tissue mixtures as well as inter-patient and intra-tissue variations. Since the state-of-the-art HLUT-even after cohort-specific optimization-cannot fully consider the broad tissue variability, patient-specific DECT-based stopping-power prediction is advisable in particle therapy.

On the equivalence of image-based dual-energy CT methods for the determination of electron density and effective atomic number in radiotherapy.

Dual-energy computed tomography enables the determination of relative electron density and effective atomic number. As this can increase accuracy in radiotherapy treatment planning, a substantial number of algorithms for the determination of the two quantities has been suggested - most of them based on reconstructed CT images. We show that many of these methods share a common theoretical framework. Equations can be transformed from one method to the other by re-definition of the calibration parameters. We suggest that further work should be spent on practical calibration and the reliability of CT numbers rather than on the theoretical framework.

Experimental verification of stopping-power prediction from single- and dual-energy computed tomography in biological tissues.

An experimental setup for consecutive measurement of ion and x-ray absorption in tissue or other materials is introduced. With this setup using a 3D-printed sample container, the reference stopping-power ratio (SPR) of materials can be measured with an uncertainty of below 0.1%. A total of 65 porcine and bovine tissue samples were prepared for measurement, comprising five samples each of 13 tissue types representing about 80% of the total body mass (three different muscle and fatty tissues, liver, kidney, brain, heart, blood, lung and bone). Using a standard stoichiometric calibration for single-energy CT (SECT) as well as a state-of-the-art dual-energy CT (DECT) approach, SPR was predicted for all tissues and then compared to the measured reference. With the SECT approach, the SPRs of all tissues were predicted with a mean error of (-0.84 ± 0.12)% and a mean absolute error of (1.27 ± 0.12)%. In contrast, the DECT-based SPR predictions were overall consistent with the measured reference with a mean error of (-0.02 ± 0.15)% and a mean absolute error of (0.10 ± 0.15)%. Thus, in this study, the potential of DECT to decrease range uncertainty could be confirmed in biological tissue.

Evaluation of Stopping-Power Prediction by Dual- and Single-Energy Computed Tomography in an Anthropomorphic Ground-Truth Phantom.

PURPOSE: To determine the accuracy of particle range prediction for proton and heavier ion radiation therapy based on dual-energy computed tomography (DECT) in a realistic inhomogeneous geometry and to compare it with the state-of-the-art clinical approach. METHODS AND MATERIALS: A 3-dimensional ground-truth map of stopping-power ratios (SPRs) was created for an anthropomorphic head phantom by assigning measured SPR values to segmented structures in a high-resolution CT scan. This reference map was validated independently comparing proton transmission measurements with Monte Carlo transport simulations. Two DECT-based methods for direct SPR prediction via the Bethe formula (DirectSPR) and 2 established approaches based on Hounsfield look-up tables (HLUTs) were chosen for evaluation. The SPR predictions from the 4 investigated methods were compared with the reference, using material-specific voxel statistics and 2-dimensional gamma analysis. Furthermore, range deviations were analyzed in an exemplary proton treatment plan. RESULTS: The established reference SPR map was successfully validated for the discrimination of SPR and range differences well below 0.3% and 1 mm, respectively, even in complex inhomogeneous settings. For the phantom materials of larger volume (mainly brain, soft tissue), the investigated methods were overall able to predict SPR within 1% median deviation. The DirectSPR methods generally performed better than the HLUT approaches. For smaller phantom parts (such as cortical bone, air cavities), all methods were affected by image smoothing, leading to considerable SPR under- or overestimation. This effect was superimposed on the general SPR prediction accuracy in the exemplary treatment plan. CONCLUSIONS: DirectSPR predictions proved to be more robust, with high accuracy in particular for larger volumes. In contrast, HLUT approaches exhibited a fortuitous component. The evaluation of accuracy in a realistic phantom with validated ground-truth SPR represents a crucial step toward possible clinical application of DECT-based SPR prediction methods.

Dual-energy CT based proton range prediction in head and pelvic tumor patients.

BACKGROUND AND PURPOSE: To reduce range uncertainty in particle therapy, an accurate computation of stopping-power ratios (SPRs) based on computed tomography (CT) is crucial. Here, we assess range differences between the state-of-the-art CT-number-to-SPR conversion using a generic Hounsfield look-up table (HLUT) and a direct patient-specific SPR prediction (RhoSigma) based on dual-energy CT (DECT) in 100 proton treatment fields. MATERIAL AND METHODS: For 25 head-tumor and 25 prostate-cancer patients, the clinically applied treatment plan, optimized using a HLUT, was recalculated with RhoSigma as CT-number-to-SPR conversion. Depth-dose curves in beam direction were extracted for both dose distributions in a regular grid and range deviations were determined and correlated to SPR differences within the irradiated volume. RESULTS: Absolute (relative) mean water-equivalent range shifts of 1.1mm (1.2%) and 4.1mm (1.7%) were observed in the head-tumor and prostate-cancer cohort, respectively. Due to the case dependency of a generic HLUT, range deviations within treatment fields strongly depend on the tissues traversed leading to a larger variation within one patient than between patients. CONCLUSIONS: The magnitude of patient-specific range deviations between HLUT and the more accurate DECT-based SPR prediction is clinically relevant. A clinical application of the latter seems feasible as demonstrated in this study using medically approved systems from CT acquisition to treatment planning.

Methodological accuracy of image-based electron density assessment using dual-energy computed tomography.

PURPOSE: Electron density is the most important tissue property influencing photon and ion dose distributions in radiotherapy patients. Dual-energy computed tomography (DECT) enables the determination of electron density by combining the information on photon attenuation obtained at two different effective x-ray energy spectra. Most algorithms suggested so far use the CT numbers provided after image reconstruction as input parameters, i.e., are imaged-based. To explore the accuracy that can be achieved with these approaches, we quantify the intrinsic methodological and calibration uncertainty of the seemingly simplest approach. METHODS: In the studied approach, electron density is calculated with a one-parametric linear superposition ('alpha blending') of the two DECT images, which is shown to be equivalent to an affine relation between the photon attenuation cross sections of the two x-ray energy spectra. We propose to use the latter relation for empirical calibration of the spectrum-dependent blending parameter. For a conclusive assessment of the electron density uncertainty, we chose to isolate the purely methodological uncertainty component from CT-related effects such as noise and beam hardening. RESULTS: Analyzing calculated spectrally weighted attenuation coefficients, we find universal applicability of the investigated approach to arbitrary mixtures of human tissue with an upper limit of the methodological uncertainty component of 0.2%, excluding high-Z elements such as iodine. The proposed calibration procedure is bias-free and straightforward to perform using standard equipment. Testing the calibration on five published data sets, we obtain very small differences in the calibration result in spite of different experimental setups and CT protocols used. Employing a general calibration per scanner type and voltage combination is thus conceivable. CONCLUSION: Given the high suitability for clinical application of the alpha-blending approach in combination with a very small methodological uncertainty, we conclude that further refinement of image-based DECT-algorithms for electron density assessment is not advisable.

Clinical Implementation of Dual-energy CT for Proton Treatment Planning on Pseudo-monoenergetic CT scans.

PURPOSE: To determine whether a standardized clinical application of dual-energy computed tomography (DECT) for proton treatment planning based on pseudomonoenergetic CT scans (MonoCTs) is feasible and increases the precision of proton therapy in comparison with single-energy CT (SECT). METHODS AND MATERIALS: To define an optimized DECT protocol, CT scan settings were analyzed experimentally concerning beam hardening, image quality, and influence on the heuristic conversion of CT numbers into stopping-power ratios (SPRs) and were compared with SECT scans with identical CT dose. Differences in range prediction and dose distribution between SECT and MonoCT were quantified for phantoms and a patient. RESULTS: Dose distributions planned on SECT and MonoCT datasets revealed mean range deviations of 0.3 mm, γ passing rates (1%, 1 mm) greater than 99.9%, and no clinically relevant changes in dose-volume histograms. However, image noise and CT-related uncertainties could be reduced by MonoCT compared with SECT, which resulted in a slightly decreased dependence of SPR prediction on beam hardening. Consequently, DECT was clinically implemented at the University Proton Therapy Dresden in 2015. Until October 2016, 150 patients were treated based on MonoCTs, and more than 950 DECT scans of 351 patients were acquired during radiation therapy. CONCLUSIONS: A standardized clinical use of MonoCT for treatment planning is feasible, leads to improved image quality and SPR prediction, extends diagnostic variety, and enables a stepwise clinical implementation of DECT toward a physics-based, patient-specific, nonheuristic SPR determination. Further reductions of CT-related uncertainties, as expected from such SPR approaches, can be evaluated on the resulting DECT patient database.

Range prediction for tissue mixtures based on dual-energy CT.

The use of dual-energy CT (DECT) potentially decreases range uncertainties in proton and ion therapy treatment planning via determination of the involved physical target quantities. For eventual clinical application, the correct treatment of tissue mixtures and heterogeneities is an essential feature, as they naturally occur within a patient's CT. Here, we present how existing methods for DECT-based ion-range prediction can be modified in order to incorporate proper mixing behavior on several structural levels. Our approach is based on the factorization of the stopping-power ratio into the relative electron density and the relative stopping number. The latter is confined for tissue between about 0.95 and 1.02 at a therapeutic beam energy of 200 MeV u(-1) and depends on the I-value. We show that convenient mixing and averaging properties arise by relating the relative stopping number to the relative cross section obtained by DECT. From this, a maximum uncertainty of the stopping-power ratio prediction below [Formula: see text] is suggested for arbitrary mixtures of human body tissues.

Technical Note: Radiological properties of tissue surrogates used in a multimodality deformable pelvic phantom for MR-guided radiotherapy.

PURPOSE: Phantom surrogates were developed to allow multimodal [computed tomography (CT), magnetic resonance imaging (MRI), and teletherapy] and anthropomorphic tissue simulation as well as materials and methods to construct deformable organ shapes and anthropomorphic bone models. METHODS: Agarose gels of variable concentrations and loadings were investigated to simulate various soft tissue types. Oils, fats, and Vaseline were investigated as surrogates for adipose tissue and bone marrow. Anthropomorphic shapes of bone and organs were realized using 3D-printing techniques based on segmentations of patient CT-scans. All materials were characterized in dual energy CT and MRI to adapt CT numbers, electron density, effective atomic number, as well as T1- and T2-relaxation times to patient and literature values. RESULTS: Soft tissue simulation could be achieved with agarose gels in combination with a gadolinium-based contrast agent and NaF to simulate muscle, prostate, and tumor tissues. Vegetable oils were shown to be a good representation for adipose tissue in all modalities. Inner bone was realized using a mixture of Vaseline and K2HPO4, resulting in both a fatty bone marrow signal in MRI and inhomogeneous areas of low and high attenuation in CT. The high attenuation of outer bone was additionally adapted by applying gypsum bandages to the 3D-printed hollow bone case with values up to 1200 HU. Deformable hollow organs were manufactured using silicone. Signal loss in the MR images based on the conductivity of the gels needs to be further investigated. CONCLUSIONS: The presented surrogates and techniques allow the customized construction of multimodality, anthropomorphic, and deformable phantoms as exemplarily shown for a pelvic phantom, which is intended to study adaptive treatment scenarios in MR-guided radiation therapy.