Identifying the optimal deep learning architecture and parameters for automatic beam aperture definition in 3D radiotherapy.

PURPOSE: Two-dimensional radiotherapy is often used to treat cervical cancer in low- and middle-income countries, but treatment planning can be challenging and time-consuming. Neural networks offer the potential to greatly decrease planning time through automation, but the impact of the wide range of hyperparameters to be set during training on model accuracy has not been exhaustively investigated. In the current study, we evaluated the effect of several convolutional neural network architectures and hyperparameters on 2D radiotherapy treatment field delineation. METHODS: Six commonly used deep learning architectures were trained to delineate four-field box apertures on digitally reconstructed radiographs for cervical cancer radiotherapy. A comprehensive search of optimal hyperparameters for all models was conducted by varying the initial learning rate, image normalization methods, and (when appropriate) convolutional kernel size, the number of learnable parameters via network depth and the number of feature maps per convolution, and nonlinear activation functions. This yielded over 1700 unique models, which were all trained until performance converged and then tested on a separate dataset. RESULTS: Of all hyperparameters, the choice of initial learning rate was most consistently significant for improved performance on the test set, with all top-performing models using learning rates of 0.0001. The optimal image normalization was not consistent across architectures. High overlap (mean Dice similarity coefficient = 0.98) and surface distance agreement (mean surface distance < 2 mm) were achieved between the treatment field apertures for all architectures using the identified best hyperparameters. Overlap Dice similarity coefficient (DSC) and distance metrics (mean surface distance and Hausdorff distance) indicated that DeepLabv3+ and D-LinkNet architectures were least sensitive to initial hyperparameter selection. CONCLUSION: DeepLabv3+ and D-LinkNet are most robust to initial hyperparameter selection. Learning rate, nonlinear activation function, and kernel size are also important hyperparameters for improving performance.

A snapshot of medical physics practice patterns.

A large number of surveys have been sent to the medical physics community addressing many clinical topics for which the medical physicist is, or may be, responsible. Each survey provides an insight into clinical practice relevant to the medical physics community. The goal of this study was to create a summary of these surveys giving a snapshot of clinical practice patterns. Surveys used in this study were created using SurveyMonkey and distributed between February 6, 2013 and January 2, 2018 via the MEDPHYS and MEDDOS listserv groups. The format of the surveys included questions that were multiple choice and free response. Surveys were included in this analysis if they met the following criteria: more than 20 responses, relevant to radiation therapy physics practice, not single-vendor specific, and formatted as multiple-choice questions (i.e., not exclusively free-text responses). Although the results of free response questions were not explicitly reported, they were carefully reviewed, and the responses were considered in the discussion of each topic. Two-hundred and fifty-two surveys were available, of which 139 passed the inclusion criteria. The mean number of questions per survey was 4. The mean number of respondents per survey was 63. Summaries were made for the following topics: simulation, treatment planning, electron treatments, linac commissioning and quality assurance, setup and treatment verification, IMRT and VMAT treatments, SRS/SBRT, breast treatments, prostate treatments, brachytherapy, TBI, facial lesion treatments, clinical workflow, and after-hours/emergent treatments. We have provided a coherent overview of medical physics practice according to surveys conducted over the last 5 yr, which will be instructive for medical physicists.

Retrospective Validation and Clinical Implementation of Automated Contouring of Organs at Risk in the Head and Neck: A Step Toward Automated Radiation Treatment Planning for Low- and Middle-Income Countries.

Purpose We assessed automated contouring of normal structures for patients with head-and-neck cancer (HNC) using a multiatlas deformable-image-registration algorithm to better provide a fully automated radiation treatment planning solution for low- and middle-income countries, provide quantitative analysis, and determine acceptability worldwide. Methods Autocontours of eight normal structures (brain, brainstem, cochleae, eyes, lungs, mandible, parotid glands, and spinal cord) from 128 patients with HNC were retrospectively scored by a dedicated HNC radiation oncologist. Contours from a 10-patient subset were evaluated by five additional radiation oncologists from international partner institutions, and interphysician variability was assessed. Quantitative agreement of autocontours with independently physician-drawn structures was assessed using the Dice similarity coefficient and mean surface and Hausdorff distances. Automated contouring was then implemented clinically and has been used for 166 patients, and contours were quantitatively compared with the physician-edited autocontours using the same metrics. Results Retrospectively, 87% of normal structure contours were rated as acceptable for use in dose-volume-histogram-based planning without edit. Upon clinical implementation, 50% of contours were not edited for use in treatment planning. The mean (± standard deviation) Dice similarity coefficient of autocontours compared with physician-edited autocontours for parotid glands (0.92 ± 0.10), brainstem (0.95 ± 0.09), and spinal cord (0.92 ± 0.12) indicate that only minor edits were performed. The average mean surface and Hausdorff distances for all structures were less than 0.15 mm and 1.8 mm, respectively. Conclusion Automated contouring of normal structures generates reliable contours that require only minimal editing, as judged by retrospective ratings from multiple international centers and clinical integration. Autocontours are acceptable for treatment planning with no or, at most, minor edits, suggesting that automated contouring is feasible for clinical use and in the ongoing development of automated radiation treatment planning algorithms.

Cost-effective immobilization for whole brain radiation therapy.

To investigate the inter- and intra-fraction motion associated with the use of a low-cost tape immobilization technique as an alternative to thermoplastic immobilization masks for whole-brain treatments. The results of this study may be of interest to clinical staff with severely limited resources (e.g., in low-income countries) and also when treating patients who cannot tolerate standard immobilization masks. Setup reproducibility of eight healthy volunteers was assessed for two different immobilization techniques. (a) One strip of tape was placed across the volunteer's forehead and attached to the sides of the treatment table. (b) A second strip was added to the first, under the chin, and secured to the table above the volunteer's head. After initial positioning, anterior and lateral photographs were acquired. Volunteers were positioned five times with each technique to allow calculation of inter-fraction reproducibility measurements. To estimate intra-fraction reproducibility, 5-minute anterior and lateral videos were taken for each technique per volunteer. An in-house software was used to analyze the photos and videos to assess setup reproducibility. The maximum intra-fraction displacement for all volunteers was 2.8 mm. Intra-fraction motion increased with time on table. The maximum inter-fraction range of positions for all volunteers was 5.4 mm. The magnitude of inter-fraction and intra-fraction motion found using the "1-strip" and "2-strip" tape immobilization techniques was comparable to motion restrictions provided by a thermoplastic mask for whole-brain radiotherapy. The results suggest that tape-based immobilization techniques represent an economical and useful alternative to the thermoplastic mask.

Contralateral breast dose from partial breast brachytherapy.

The purpose of this study was to determine the dose to the contralateral breast during accelerated partial breast irradiation (APBI) and to compare it to external beam-published values. Thermoluminescent dosimeter (TLD) packets were used to measure the dose to the most medial aspect of the contralateral breast during APBI simulation, daily quality assurance (QA), and treatment. All patients in this study were treated with a single-entry, multicatheter device for 10 fractions to a total dose of 34 Gy. A mark was placed on the patient's skin on the medial aspect of the opposite breast. Three TLD packets were taped to this mark during the pretreatment simulation. Simulations consisted of an AP and Lateral scout and a limited axial scan encompassing the lumpectomy cavity (miniscan), if rotation was a concern. After the simulation the TLD packets were removed and the patients were moved to the high-dose-rate (HDR) vault where three new TLD packets were taped onto the patients at the skin mark. Treatment was administered with a Nucletron HDR afterloader using Iridium-192 as the treatment source. Post-treatment, TLDs were read (along with the simulation and QA TLD and a set of standards exposed to a known dose of 6 MV photons). Measurements indicate an average total dose to the contralateral breast of 70 cGy for outer quadrant implants and 181 cGy for inner quadrant implants. Compared to external beam breast tangents, these results point to less dose being delivered to the contralateral breast when using APBI.

Commissioning results of an automated treatment planning verification system.

A dose calculation verification system (VS) was acquired and commissioned as a second check on the treatment planning system (TPS). This system reads DICOM CT datasets, RT plans, RT structures, and RT dose from the TPS and automatically, using its own collapsed cone superposition/convolution algorithm, computes dose on the same CT dataset. The system was commissioned by extracting basic beam parameters for simple field geometries and dose verification for complex treatments. Percent depth doses (PDD) and profiles were extracted for field sizes using jaw settings 3 × 3 cm2 - 40 × 40 cm2 and compared to measured data, as well as our TPS model. Smaller fields of 1 × 1 cm2 and 2 × 2 cm2 generated using the multileaf collimator (MLC) were analyzed in the same fashion as the open fields. In addition, 40 patient plans consisting of both IMRT and VMAT were computed and the following comparisons were made: 1) TPS to the VS, 2) VS to measured data, and 3) TPS to measured data where measured data is both ion chamber (IC) and film measurements. Our results indicated for all field sizes using jaw settings PDD errors for the VS on average were less than 0.87%, 1.38%, and 1.07% for 6x, 15x, and 18x, respectively, relative to measured data. PDD errors for MLC field sizes were less than 2.28%, 1.02%, and 2.23% for 6x, 15x, and 18x, respectively. The infield profile analysis yielded results less than 0.58% for 6x, 0.61% for 15x, and 0.77% for 18x for the VS relative to measured data. Analysis of the penumbra region yields results ranging from 66.5% points, meeting the DTA criteria to 100% of the points for smaller field sizes for all energies. Analysis of profile data for field sizes generated using the MLC saw agreement with infield DTA analysis ranging from 68.8%-100% points passing the 1.5%/1.5 mm criteria. Results from the dose verification for IMRT and VMAT beams indicated that, on average, the ratio of TPS to IC and VS to IC measurements was 100.5 ± 1.9% and 100.4 ± 1.3%, respectively, while our TPS to VS was 100.1 ± 1.0%. When comparing the TPS and VS to film measurements, the average percentage pixels passing a 3%/3mm criteria based gamma analysis were 96.6 ± 4.2% and 97 ± 5.6%, respectively. When the VS was compared to the TPS, on average 98.1 ± 5.3% of pixels passed the gamma analysis. Based upon these preliminary results, the VS system should be able to calculate dose adequately as a verification tool of our TPS.

Impact of heterogeneity-based dose calculation using a deterministic grid-based Boltzmann equation solver for intracavitary brachytherapy.

PURPOSE: To investigate the dosimetric impact of the heterogeneity dose calculation Acuros (Transpire Inc., Gig Harbor, WA), a grid-based Boltzmann equation solver (GBBS), for brachytherapy in a cohort of cervical cancer patients. METHODS AND MATERIALS: The impact of heterogeneities was retrospectively assessed in treatment plans for 26 patients who had previously received (192)Ir intracavitary brachytherapy for cervical cancer with computed tomography (CT)/magnetic resonance-compatible tandems and unshielded colpostats. The GBBS models sources, patient boundaries, applicators, and tissue heterogeneities. Multiple GBBS calculations were performed with and without solid model applicator, with and without overriding the patient contour to 1 g/cm(3) muscle, and with and without overriding contrast materials to muscle or 2.25 g/cm(3) bone. Impact of source and boundary modeling, applicator, tissue heterogeneities, and sensitivity of CT-to-material mapping of contrast were derived from the multiple calculations. American Association of Physicists in Medicine Task Group 43 (TG-43) guidelines and the GBBS were compared for the following clinical dosimetric parameters: Manchester points A and B, International Commission on Radiation Units and Measurements (ICRU) report 38 rectal and bladder points, three and nine o'clock, and (D2cm3) to the bladder, rectum, and sigmoid. RESULTS: Points A and B, D(2) cm(3) bladder, ICRU bladder, and three and nine o'clock were within 5% of TG-43 for all GBBS calculations. The source and boundary and applicator account for most of the differences between the GBBS and TG-43 guidelines. The D(2cm3) rectum (n = 3), D(2cm3) sigmoid (n = 1), and ICRU rectum (n = 6) had differences of >5% from TG-43 for the worst case incorrect mapping of contrast to bone. Clinical dosimetric parameters were within 5% of TG-43 when rectal and balloon contrast were mapped to bone and radiopaque packing was not overridden. CONCLUSIONS: The GBBS has minimal impact on clinical parameters for this cohort of patients with unshielded applicators. The incorrect mapping of rectal and balloon contrast does not have a significant impact on clinical parameters. Rectal parameters may be sensitive to the mapping of radiopaque packing.

On the feasibility of treating to a 1.5 cm PTV with a commercial single-entry hybrid applicator in APBI breast brachytherapy.

PURPOSE: To evaluate and determine whether 30 patients previously treated with the SAVI™ device could have been treated to a PTV_EVAL created with a 1.5 cm expansion. This determination was based upon dosimetric parameters derived from current recommendations and dose-response data. MATERIAL AND METHODS: Thirty patients were retrospectively planned with PTV_EVALs generated with a 1.5 cm expansion (PTV_EVAL_1.5). Plans were evaluated based on PTV_EVAL_1.5 coverage (V90, V95, V100), skin and rib maximum doses (0.1 cc maximum dose as a percentage of prescription dose), as well as V150 and V200 for the PTV_EVAL_1.5. The treatment planning goal was to deliver ≥90% of the prescribed dose to ≥90% of the PTV_EVAL_1.5. Skin and rib maximum doses were to be ≤125% of the prescription dose and preferably ≤100% of the prescription dose. V150 and V200 were not allowed to exceed 52.5 cc and 21 cc, respectively. Plans not meeting the above criteria were recomputed with a 1.25 cm expanded PTV_EVAL and re-evaluated. RESULTS: Based on the above dose constraints, 30% (9/30) of the patients evaluated could have been treated with a 1.5 cm PTV_EVAL. The breakdown of cases successfully achieving the above dose constraints by applicator was: 0/4 (0%) 6-1, 6/15 (40%) 8-1, and 3/11 (27%) 10-1. For these PTV_EVAL_1.5 plans, median V90% was 90.3%, whereas the maximum skin and rib doses were all less than 115.2% and 117.6%, respectively. The median V150 and V200 volumes were 39.2 cc and 19.3, respectively. The treated PTV_EVAL_1.5 was greater in volume than the PTV_EVAL by 41.7 cc, and 60 cc for the 8-1, and 10-1 applicators, respectively. All remaining plans (17) successfully met the above dose constraints to be treated with a 1.25 cm PTV_EVAL (PTV_EVAL_1.25). For the PTV_EVAL_1.25 plans, V90% was 93.7%, and the maximum skin and rib doses were all less than 109.2% and 102.5%, respectively. The median V150 and V200 volumes were 41.2 cc and 19.3, respectively. The treated PTV_EVAL_1.25 was greater in volume than the PTV_EVAL by 16 cc, 24.9 cc, and 33.5 cc for the 6-1, 8-1 and 10-1 applicators, respectively. CONCLUSIONS: It is dosimetrically possible to treat beyond the currently advised 1.0 cm expanded PTV_EVAL. Most patients should be able to be treated with a 1.25 cm PTV_EVAL and a select group with a 1.5 cm PTV_EVAL. Applicator size appears to determine the ability to expand to a 1.5 cm PTV_EVAL, as smaller devices were not as propitious in this regard. Further studies may identify additional patient groups that would benefit from this approach.