Human factors in the clinical implementation of deep learning-based automated contouring of pelvic organs at risk for MRI-guided radiotherapy.

PURPOSE: Deep neural nets have revolutionized the science of auto-segmentation and present great promise for treatment planning automation. However, little data exists regarding clinical implementation and human factors. We evaluated the performance and clinical implementation of a novel deep learning-based auto-contouring workflow for 0.35T magnetic resonance imaging (MRI)-guided pelvic radiotherapy, focusing on automation bias and objective measures of workflow savings. METHODS: An auto-contouring model was developed using a UNet-derived architecture for the femoral heads, bladder, and rectum in 0.35T MR images. Training data was taken from 75 patients treated with MRI-guided radiotherapy at our institution. The model was tested against 20 retrospective cases outside the training set, and subsequently was clinically implemented. Usability was evaluated on the first 30 clinical cases by computing Dice coefficient (DSC), Hausdorff distance (HD), and the fraction of slices that were used un-modified by planners. Final contours were retrospectively reviewed by an experienced planner and clinical significance of deviations was graded as negligible, low, moderate, and high probability of leading to actionable dosimetric variations. In order to assess whether the use of auto-contouring led to final contours more or less in agreement with an objective standard, 10 pre-treatment and 10 post-treatment blinded cases were re-contoured from scratch by three expert planners to get expert consensus contours (EC). EC was compared to clinically used (CU) contours using DSC. Student's t-test and Levene's statistic were used to test statistical significance of differences in mean and standard deviation, respectively. Finally, the dosimetric significance of the contour differences were assessed by comparing the difference in bladder and rectum maximum point doses between EC and CU before and after the introduction of automation. RESULTS: Median (interquartile range) DSC for the retrospective test data were 0.92(0.02), 0.92(0.06), 0.93(0.06), 0.87(0.04) for the post-processed contours for the right and left femoral heads, bladder, and rectum, respectively. Post-implementation median DSC were 1.0(0.0), 1.0(0.0), 0.98(0.04), and 0.98(0.06), respectively. For each organ, 96.2, 95.4, 59.5, and 68.21 percent of slices were used unmodified by the planner. DSC between EC and pre-implementation CU contours were 0.91(0.05*), 0.91*(0.05*), 0.95(0.04), and 0.88(0.04) for right and left femoral heads, bladder, and rectum, respectively. The corresponding DSC for post-implementation CU contours were 0.93(0.02*), 0.93*(0.01*), 0.96(0.01), and 0.85(0.02) (asterisks indicate statistically significant difference). In a retrospective review of contours used for planning, a total of four deviating slices in two patients were graded as low potential clinical significance. No deviations were graded as moderate or high. Mean differences between EC and CU rectum max-doses were 0.1 ± 2.6 Gy and -0.9 ± 2.5 Gy for pre- and post-implementation, respectively. Mean differences between EC and CU bladder/bladder wall max-doses were -0.9 ± 4.1 Gy and 0.0 ± 0.6 Gy for pre- and post-implementation, respectively. These differences were not statistically significant according to Student's t-test. CONCLUSION: We have presented an analysis of the clinical implementation of a novel auto-contouring workflow. Substantial workflow savings were obtained. The introduction of auto-contouring into the clinical workflow changed the contouring behavior of planners. Automation bias was observed, but it had little deleterious effect on treatment planning.

Ablative radiotherapy for liver tumors using stereotactic MRI-guidance: A prospective phase I trial.

BACKGROUND AND PURPOSE: To prospectively determine the feasibility, safety, and efficacy of stereotactic body radiation therapy (SBRT) to primary and secondary liver tumors with MR-guided radiation therapy (MRgRT). MATERIALS AND METHODS: Treatment plans with a conventional CT-guided linear accelerator and a MRI-guided tri-60Co teletherapy unit (MR-Co) were generated and compared for patients undergoing liver-directed SBRT from 2015 to 2017. If dosimetric parameters were met on MR-Co, patients were treated with MRgRT. The highest priority constraint was >1000 cc or >800 cc of normal liver receiving <15 Gy for single- or multiple-lesion treatments, respectively. Treatment was delivered every other day. RESULTS: Of 23 patients screened, 20 patients (8 primary, 12 secondary) and 25 liver tumors underwent MR-guided SBRT to a median dose of 54 Gy (range 11.5-60) in a median of 3 fractions (range 1-5). With a median follow up of 18.9 months, the 1- and 2-year estimate of local control were 94.7% and 79.6%, respectively. A difference in local control between single and multiple lesions or BED ≥ 100 Gy10 and BED < 100 Gy10, respectively, was observed. The 2-year estimate of overall survival (OS) was 50.7% with a median OS of 29 months. There were no acute grade ≥ 3 toxicities and one late grade 3/4 toxicity from a single patient whose plan exceeded an unrecognized dose constraint at the time. CONCLUSION: MR-guided SBRT is a viable and safe option in the delivery of ultrahypofractionated ablative radiation treatment to primary and secondary liver tumors resulting in high rates of local control and very favorable toxicity profiles.

A generative adversarial network-based (GAN-based) architecture for automatic fiducial marker detection in prostate MRI-only radiotherapy simulation images.

PURPOSE: Clinical sites utilizing magnetic resonance imaging (MRI)-only simulation for prostate radiotherapy planning typically use fiducial markers for pretreatment patient positioning and alignment. Fiducial markers appear as small signal voids in MRI images and are often difficult to discern. Existing clinical methods for fiducial marker localization require multiple MRI sequences and/or manual interaction and specialized expertise. In this study, we develop a robust method for automatic fiducial marker detection in prostate MRI simulation images and quantify the pretreatment alignment accuracy using automatically detected fiducial markers in MRI. METHODS AND MATERIALS: In this study, a deep learning-based algorithm was used to convert MRI images into labeled fiducial marker volumes. Seventy-seven prostate cancer patients who received marker implantation prior to MRI and CT simulation imaging were selected for this study. Multiple-Echo T1 -VIBE MRI images were acquired, and images were stratified (at the patient level) based on the presence of intraprostatic calcifications. Ground truth (GT) contours were defined by an expert on MRI using CT images. Training was done using the pix2pix generative adversarial network (GAN) image-to-image translation package and model testing was done using fivefold cross validation. For performance comparison, an experienced medical dosimetrist and a medical physicist each manually contoured fiducial markers in MRI images. The percent of correct detections and F1 classification scores are reported for markers detected using the automatic detection algorithm and human observers. The patient positioning errors were quantified by calculating the target registration errors (TREs) from fiducial marker driven rigid registration between MRI and CBCT images. Target registration errors were quantified for fiducial marker contours defined on MRI by the automatic detection algorithm and the two expert human observers. RESULTS: Ninety-six percent of implanted fiducial markers were correctly identified using the automatic detection algorithm. Two expert raters correctly identified 97% and 96% of fiducial markers, respectively. The F1 classification score was 0.68, 0.75, and 0.72 for the automatic detection algorithm and two human raters, respectively. The main source of false discoveries was intraprostatic calcifications. The mean TRE differences between alignments from automatic detection algorithm and human detected markers and GT were <1 mm. CONCLUSIONS: We have developed a deep learning-based approach to automatically detect fiducial markers in MRI-only simulation images in a clinically representative patient cohort. The automatic detection algorithm-predicted markers can allow for patient setup with similar accuracy to independent human observers.

Hazards of sparing the ipsilateral parotid gland in the node-positive neck with intensity modulated radiation therapy: Spatial analysis of regional recurrence risk.

PURPOSE: The practice of deliberately sparing the ipsilateral parotid gland with intensity modulated radiation therapy (IMRT) in patients with node-positive head and neck cancer is controversial. We sought to compare the clinical outcomes among consecutive cohorts of patients with head and neck cancer who were treated with differing strategies to spare the parotid gland that is ipsilateral to the involved neck using IMRT. METHODS AND MATERIALS: A total of 305 patients were treated with IMRT for node-positive squamous cell carcinoma of the head and neck. The first 139 patients were treated with IMRT whereby the ipsilateral parotid gland was delineated and intentionally designated as an avoidance structure during planning. The subsequent 166 patients were treated by IMRT without the deliberate sparing of the ipsilateral parotid gland. RESULTS: The 2-year estimates of overall survival, local-regional control, and distant metastasis-free survival were 84%, 73%, and 87%, respectively. The 2-year estimates of overall survival were 77% and 86% among patients who were treated by IMRT with and without the sparing of the ipsilateral parotid gland, respectively (P = .01). The respective rates of 2-year regional control were 76% and 90% (P < .001). A trend was observed between increased nodal burden in the ipsilateral cervical neck and the likelihood of regional failure for both groups. A spatial evaluation revealed a significantly higher incidence of marginal failures and true misses in the cohort of patients who underwent IMRT with the sparing of the ipsilateral parotid gland. CONCLUSION: Caution is urged when using IMRT to spare patients' parotid gland on the involved side of neck disease. Our study showed a significantly higher preponderance of regional failure, which highlights the need for careful patient selection and consideration of clinical and pathological factors that influence the likelihood of disease recurrence in the ipsilateral neck.

Magnetic resonance imaging guided reirradiation of recurrent and second primary head and neck cancer.

PURPOSE: To report a single-institutional experience using magnetic resonance imaging (MRI) guided radiation therapy for the reirradiation of recurrent and second cancers of the head and neck. METHODS AND MATERIALS: Between October 2014 and August 2016, 13 consecutive patients with recurrent or new primary cancers of the head and neck that occurred in a previously irradiated field were prospectively enrolled in an institutional registry trial to investigate the feasibility and efficacy of MRI guided radiation therapy using a 0.35-T MRI scanner with a cobalt-60 radiation therapy source called the ViewRay system (ViewRay Inc., Cleveland, OH). Eligibility criteria included biopsy-proven evidence of recurrent or new primary squamous cell carcinoma of the head and neck, measurable disease, and previous radiation to >60 Gy. MRI guided reirradiation was delivered either using intensity modulated radiation therapy with conventional fractionation to a median dose of 66 Gy or stereotactic body radiation therapy (SBRT) using 7 to 8 Gy fractions on nonconsecutive days to a median dose of 40 Gy. Two patients (17%) received concurrent chemotherapy. RESULTS: The 1- and 2-year estimates of in-field control were 72% and 72%, respectively. A total of 227 daily MRI scans were obtained to guide reirradiation. The 2-year estimates of overall survival and progression-free survival were 53% and 59%, respectively. There were no treatment-related fatalities or hospitalizations. Complications included skin desquamation, odynophagia, otitis externa, keratitis and/or conjunctivitis, and 1 case of aspiration pneumonia. CONCLUSIONS: Our preliminary findings show that reirradiation with MRI guided radiation therapy results in effective disease control with relatively low morbidity for patients with recurrent and second primary cancers of the head and neck. The superior soft tissue resolution of the MRI scans that were used for planning and delivery has the potential to improve the therapeutic ratio.

Tolerance of the Brachial Plexus to High-Dose Reirradiation.

PURPOSE: To study the tolerance of the brachial plexus to high doses of radiation exceeding historically accepted limits by analyzing human subjects treated with reirradiation for recurrent tumors of the head and neck. METHODS AND MATERIALS: Data from 43 patients who were confirmed to have received overlapping dose to the brachial plexus after review of radiation treatment plans from the initial and reirradiation courses were used to model the tolerance of this normal tissue structure. A standardized instrument for symptoms of neuropathy believed to be related to brachial plexus injury was utilized to screen for toxicity. Cumulative dose was calculated by fusing the initial dose distributions onto the reirradiation plan, thereby creating a composite plan via deformable image registration. The median elapsed time from the initial course of radiation therapy to reirradiation was 24 months (range, 3-144 months). RESULTS: The dominant complaints among patients with symptoms were ipsilateral pain (54%), numbness/tingling (31%), and motor weakness and/or difficulty with manual dexterity (15%). The cumulative maximum dose (Dmax) received by the brachial plexus ranged from 60.5 Gy to 150.1 Gy (median, 95.0 Gy). The cumulative mean (Dmean) dose ranged from 20.2 Gy to 111.5 Gy (median, 63.8 Gy). The 1-year freedom from brachial plexus-related neuropathy was 67% and 86% for subjects with a cumulative Dmax greater than and less than 95.0 Gy, respectively (P=.05). The 1-year complication-free rate was 66% and 87%, for those reirradiated within and after 2 years from the initial course, respectively (P=.06). CONCLUSION: The development of brachial plexus-related symptoms was less than expected owing to repair kinetics and to the relatively short survival of the subject population. Time-dose factors were demonstrated to be predictive of complications.

Inadequate target volume delineation and local-regional recurrence after intensity-modulated radiotherapy for human papillomavirus-positive oropharynx cancer.

PURPOSE: To perform a spatial analysis of local-regional recurrences, in relation to quantitative dose distribution, among patients treated by intensity-modulated radiotherapy (IMRT) for human papillomavirus (HPV)-positive oropharyngeal cancer. METHODS AND MATERIALS: The records of 107 consecutive patients who presented for consideration of re-irradiation for local-regional recurrent disease occurring in a previously irradiated field were reviewed. The original IMRT plans were retrieved for those with HPV-positive disease originating from the oropharynx, and deformable image registration was used to fuse the magnetic resonance imaging (MRI) and positron emission tomography (PET) scans obtained at recurrence to the pre-treatment planning computed tomography (CT) dataset. The recurrent tumor volume (Vrecur) was subsequently identified on axial imaging, and the dose of radiation received by Vrecur was then calculated and analyzed using dose-volume histograms. RESULTS: A total of 83 recurrent lesions occurring in 50 oropharyngeal cancer patients were HPV-positive and met inclusion criteria. Using PET-defined Vrecur, thirty-three lesions were classified as in-field recurrences (40%), 35 were marginal misses (41%), and 15 were true misses (18%). Using the MRI-defined Vrecur, thirty-seven lesions were classified as in-field recurrences (45%), 32 were marginal misses (39%), and 14 were true misses (17%). CONCLUSION: A significant proportion of local-regional recurrences from HPV-positive oropharyngeal cancer represented geographical misses which possibly could have been prevented with more meticulous attention to IMRT planning. This finding has important implications with respect to ongoing attempts to de-escalate radiation dose for this disease. Our data highlight the importance of robust quality assurance with careful review of target volumes prior to the initiation of IMRT.

Feasibility of magnetic resonance imaging-guided liver stereotactic body radiation therapy: A comparison between modulated tri-cobalt-60 teletherapy and linear accelerator-based intensity modulated radiation therapy.

PURPOSE: The purpose of this study was to investigate the dosimetric feasibility of liver stereotactic body radiation therapy (SBRT) using a teletherapy system equipped with 3 rotating (60)Co sources (tri-(60)Co system) and a built-in magnetic resonance imager (MRI). We hypothesized tumor size and location would be predictive of favorable dosimetry with tri-(60)Co SBRT. METHODS AND MATERIALS: The primary study population consisted of 11 patients treated with SBRT for malignant hepatic lesions whose linear accelerator (LINAC)-based SBRT plans met all mandatory Radiation Therapy Oncology Group (RTOG) 1112 organ-at-risk (OAR) constraints. The secondary study population included 5 additional patients whose plans did not meet the mandatory constraints. Patients received 36 to 60 Gy in 3 to 5 fractions. Tri-(60)Co system SBRT plans were planned with ViewRay system software. RESULTS: All patients in the primary study population had tri-(60)Co SBRT plans that passed all RTOG constraints, with similar planning target volume coverage and OAR doses to LINAC plans. Mean liver doses and V10Gy to the liver, although easily meeting RTOG 1112 guidelines, were significantly higher with tri-(60)Co plans. When the 5 additional patients were included in a univariate analysis, the tri-(60)Co SBRT plans were still equally able to pass RTOG constraints, although they did have inferior ability to pass more stringent liver and kidney constraints (P < .05). A multivariate analysis found the ability of a tri-(60)Co SBRT plan to meet these constraints depended on lesion location and size. Patients with smaller or more peripheral lesions (as defined by distance from the aorta, chest wall, liver dome, and relative lesion volume) were significantly more likely to have tri-(60)Co plans that spared the liver and kidney as well as LINAC plans did (P < .05). CONCLUSIONS: It is dosimetrically feasible to perform liver SBRT with a tri-(60)Co system with a built-in MRI. Patients with smaller or more peripheral lesions are more likely to have optimal liver and kidney sparing, with the added benefit of MRI guidance, when receiving tri-(60)Co-based SBRT.

Dosimetric feasibility of magnetic resonance imaging-guided tri-cobalt 60 preoperative intensity modulated radiation therapy for soft tissue sarcomas of the extremity.

PURPOSE: The purpose of this study was to investigate the dosimetric differences of delivering preoperative intensity modulated radiation therapy (IMRT) to patients with soft tissue sarcomas of the extremity (ESTS) with a teletherapy system equipped with 3 rotating (60)Co sources and a built-in magnetic resonance imaging and with standard linear accelerator (LINAC)-based IMRT. METHODS AND MATERIALS: The primary study population consisted of 9 patients treated with preoperative radiation for ESTS between 2008 and 2014 with LINAC-based static field IMRT. LINAC plans were designed to deliver 50 Gy in 25 fractions to 95% of the planning target volume (PTV). Tri-(60)Co system IMRT plans were designed with ViewRay system software. RESULTS: Tri-(60)Co-based IMRT plans achieved equivalent target coverage and dosimetry for organs at risk (long bone, skin, and skin corridor) compared with LINAC-based IMRT plans. The maximum and minimum PTV doses, heterogeneity indices, and ratio of the dose to 50% of the volume were equivalent for both planning systems. One LINAC plan violated the maximum bone dose constraint, whereas none of the tri-(60)Co plans did. CONCLUSIONS: Using a tri-(60)Co system, we were able to achieve equivalent dosimetry to the PTV and organs at risk for patients with ESTS compared with LINAC-based IMRT plans. The tri-(60)Co system may be advantageous over current treatment platforms by allowing PTV reduction and by elimination of the additional radiation dose associated with daily image guidance, but this needs to be evaluated prospectively.