No abstract

Purpose To reduce workload and inconsistencies in organ segmentation for radiation treatment planning, we developed and evaluated general and custom autosegmentation models on computed tomography (CT) for three major tumor sites using a well‐established deep convolutional neural network (DCNN). Methods Five CT‐based autosegmentation models for 42 organs at risk (OARs) in head and neck (HN), abdomen (ABD), and male pelvis (MP) were developed using a full three‐dimensional (3D) DCNN architecture. Two types of deep learning (DL) models were separately trained using either general diversified multi‐institutional datasets or custom well‐controlled single‐institution datasets. To improve segmentation accuracy, an adaptive spatial resolution approach for small and/or narrow OARs and a pseudo scan extension approach, when CT scan length is too short to cover entire organs, were implemented. The performance of the obtained models was evaluated based on accuracy and clinical applicability of the autosegmented contours using qualitative visual inspection and quantitative calculation of dice similarity coefficient (DSC), mean distance to agreement (MDA), and time efficiency. Results The five DL autosegmentation models developed for the three anatomical sites were found to have high accuracy (DSC ranging from 0.8 to 0.98) for 74% OARs and marginally acceptable for 26% OARs. The custom models performed slightly better than the general models, even with smaller custom datasets used for the custom model training. The organ‐based approaches improved autosegmentation accuracy for small or complex organs (e.g., eye lens, optic nerves, inner ears, and bowels). Compared with traditional manual contouring times, the autosegmentation times, including subsequent manual editing, if necessary, were substantially reduced by 88% for MP, 80% for HN, and 65% for ABD models. Conclusions The obtained autosegmentation models, incorporating organ‐based approaches, were found to be effective and accurate for most OARs in the male pelvis, head and neck, and abdomen. We have demonstrated that our multianatomical DL autosegmentation models are clinically useful for radiation treatment planning.

When asked 'How would you rate your transportation experience today?' 82% responded Above Average. To the question 'Would you have been able to attend your appointment today if this program did not exist?' 92% answered No. Conclusion: This study shows that the cost of rideshare transportation can be significantly less than the cost of no-shows. This suggests that a proactive virtual transportation hub can help address transportation barriers, drive patient satisfaction and reduce the waste of no-shows. Radiation therapy represents an ambulatory medicine crucible for patients with limited transportation and social support. Scaling up rideshare innovations from radiation oncology has the potential to drive broader ambulatory strategy.

Rapid and accurate generation of synthetic CT (sCT) from daily MRI, required in MR-guided adaptive radiotherapy (MRgART), is challenging in abdomen due to the air volumes that can change quickly and randomly (thus, no paired CT available) and hard to automatically segment on daily MRI. This work aims to develop a novel structure-preservation deep learning method to quickly create sCT from a special MRI sequence. Materials/Methods: The sCT model was based on the generative adversarial networks (GANs) technology with innovations including extra deformable layers in sub-networks and mutual information loss terms, which were added to effectively guide the network to preserve true structures from MRI for those highly deformed organs and air pockets in abdomen. The model was developed to use air scan, a specially designed MRI sequence to image air in abdomen, which a 3D FLASH Cartesian sequence that had optimized RF pulses to achieve a minimum echo time of 1 msec with heavy acceleration to keep scan times under 9 seconds to avoid motion artifacts. Daily air scans acquired along with daily plan MRI on a 1.5T MR-Linac during MRgART for 21 patients with abdominal tumors were used to create the sCT model (sCT-DL). The sCT-DL was compared with the results from: (1) sCT-DIR, generated via deformable image registration (DIR) of reference CT and daily plan MRI, and (i) sCT-Gdiff, a previously reported method to automatically segment air volumes on daily MR by a threshold in a union of (i) deformed air-containing organs (e.g., bowels) and (ii) a expansion to account for DIR inaccuracy (calculated by taking the root mean square error between primary and deformed secondary images, divided by the gradient of primary). In addition, the air volumes on sCT-DL obtained with a threshold of HU < -300 and on sCT-Gdiff were compared to those manually delineated on the air scans (Airmanual) based on Dice similarity coefficient (DSC). Dose calculated for a MR-Linac plan on sCT-DL was compared to those calculated on sCT-DIR and sCT-Gdiff. Dosimetric accuracy was measured using the fractional volume with dose disagreement < 3% (FV3). The bone volumes from sCT-DL were compared to those on sCT-DIR (ground truth) based on DSC and FV3. Results: The sCT-DL creation was very fast (< 1.0 sec with a hardware of 28 CPUs and P2000 GPU). The air volume DSC was 0.49 § 0.1 for sCT-DL and 0.89 § 0.06 for sCT-Gdiff, as compared to the Air-manual volumes. For dose accuracy, the FV3 was 0.86 § 0.03 for sCT-DL versus 0.90 § 0.01 for sCT-Gdiff. For the bone volumes on sCT-DL, DSC was 0.54 § 0.04, and FV3 was 0.87 § 0.01 as compared to the ground truth of sCT-DIR. Conclusion: It is promising to use the proposed novel structure-preservation deep learning method to automatically generated sCT in abdomen based on this proof-of-principle study. The sCT can be generated within 10 sec including the special MRI acquisition. With further development using large datasets, the novel sCT method may be implemented for MRgART of abdominal tumors.

Objective. In the current MR-Linac online adaptive workflow, air regions on the MR images need to be manually delineated for abdominal targets, and then overridden by air density for dose calculation. Auto-delineation of these regions is desirable for speed purposes, but poses a challenge, since unlike computed tomography, they do not occupy all dark regions on the image. The purpose of this study is to develop an automated method to segment the air regions on MRI-guided adaptive radiation therapy (MRgART) of abdominal tumors. Approach. A modified ResUNet3D deep learning (DL)-based auto air delineation model was trained using 102 patients’ MR images. The MR images were acquired by a dedicated in-house sequence named ‘Air-Scan’, which is designed to generate air regions that are especially dark and accentuated. The air volumes generated by the newly developed DL model were compared with the manual air contours using geometric similarity (Dice Similarity Coefficient (DSC)), and dosimetric equivalence using Gamma index and dose-volume parameters. Main results. The average DSC agreement between the DL generated and manual air contours is 99% ± 1%. The gamma index between the dose calculations with overriding the DL versus manual air volumes with density of 0.01 is 97% ± 2% for a local gamma calculation with a tolerance of 2% and 2 mm. The dosimetric parameters from planning target volume—PTV and organs at risk—OARs were all within 1% between when DL versus manual contours were overridden by air density. The model runs in less than five seconds on a PC with 28 Core processor and NVIDIA Quadro® P2000 GPU. Significance: a DL based automated segmentation method was developed to generate air volumes on specialized abdominal MR images and generate results that are practically equivalent to the manual contouring of air volumes.

No abstract

IntroductionMulti-sequence multi-parameter MRIs are often used to define targets and/or organs at risk (OAR) in radiation therapy (RT) planning. Deep learning has so far focused on developing auto-segmentation models based on a single MRI sequence. The purpose of this work is to develop a multi-sequence deep learning based auto-segmentation (mS-DLAS) based on multi-sequence abdominal MRIs.Materials and methodsUsing a previously developed 3DResUnet network, a mS-DLAS model using 4 T1 and T2 weighted MRI acquired during routine RT simulation for 71 cases with abdominal tumors was trained and tested. Strategies including data pre-processing, Z-normalization approach, and data augmentation were employed. Additional 2 sequence specific T1 weighted (T1-M) and T2 weighted (T2-M) models were trained to evaluate performance of sequence-specific DLAS. Performance of all models was quantitatively evaluated using 6 surface and volumetric accuracy metrics.ResultsThe developed DLAS models were able to generate reasonable contours of 12 upper abdomen organs within 21 seconds for each testing case. The 3D average values of dice similarity coefficient (DSC), mean distance to agreement (MDA mm), 95 percentile Hausdorff distance (HD95% mm), percent volume difference (PVD), surface DSC (sDSC), and relative added path length (rAPL mm/cc) over all organs were 0.87, 1.79, 7.43, -8.95, 0.82, and 12.25, respectively, for mS-DLAS model. Collectively, 71% of the auto-segmented contours by the three models had relatively high quality. Additionally, the obtained mS-DLAS successfully segmented 9 out of 16 MRI sequences that were not used in the model training.ConclusionWe have developed an MRI-based mS-DLAS model for auto-segmenting of upper abdominal organs on MRI. Multi-sequence segmentation is desirable in routine clinical practice of RT for accurate organ and target delineation, particularly for abdominal tumors. Our work will act as a stepping stone for acquiring fast and accurate segmentation on multi-contrast MRI and make way for MR only guided radiation therapy.

BackgroundIn order to accurately accumulate delivered dose for head and neck cancer patients treated with the Adapt to Position workflow on the 1.5T magnetic resonance imaging (MRI)‐linear accelerator (MR‐linac), the low‐resolution T2‐weighted MRIs used for daily setup must be segmented to enable reconstruction of the delivered dose at each fraction.PurposeIn this pilot study, we evaluate various autosegmentation methods for head and neck organs at risk (OARs) on on‐board setup MRIs from the MR‐linac for off‐line reconstruction of delivered dose.MethodsSeven OARs (parotid glands, submandibular glands, mandible, spinal cord, and brainstem) were contoured on 43 images by seven observers each. Ground truth contours were generated using a simultaneous truth and performance level estimation (STAPLE) algorithm. Twenty total autosegmentation methods were evaluated in ADMIRE: 1–9) atlas‐based autosegmentation using a population atlas library (PAL) of 5/10/15 patients with STAPLE, patch fusion (PF), random forest (RF) for label fusion; 10–19) autosegmentation using images from a patient's 1–4 prior fractions (individualized patient prior [IPP]) using STAPLE/PF/RF; 20) deep learning (DL) (3D ResUNet trained on 43 ground truth structure sets plus 45 contoured by one observer). Execution time was measured for each method. Autosegmented structures were compared to ground truth structures using the Dice similarity coefficient, mean surface distance (MSD), Hausdorff distance (HD), and Jaccard index (JI). For each metric and OAR, performance was compared to the inter‐observer variability using Dunn's test with control. Methods were compared pairwise using the Steel‐Dwass test for each metric pooled across all OARs. Further dosimetric analysis was performed on three high‐performing autosegmentation methods (DL, IPP with RF and 4 fractions [IPP_RF_4], IPP with 1 fraction [IPP_1]), and one low‐performing (PAL with STAPLE and 5 atlases [PAL_ST_5]). For five patients, delivered doses from clinical plans were recalculated on setup images with ground truth and autosegmented structure sets. Differences in maximum and mean dose to each structure between the ground truth and autosegmented structures were calculated and correlated with geometric metrics.ResultsDL and IPP methods performed best overall, all significantly outperforming inter‐observer variability and with no significant difference between methods in pairwise comparison. PAL methods performed worst overall; most were not significantly different from the inter‐observer variability or from each other. DL was the fastest method (33 s per case) and PAL methods the slowest (3.7–13.8 min per case). Execution time increased with a number of prior fractions/atlases for IPP and PAL. For DL, IPP_1, and IPP_RF_4, the majority (95%) of dose differences were within ± 250 cGy from ground truth, but outlier differences up to 785 cGy occurred. Dose differences were much higher for PAL_ST_5, with outlier differences up to 1920 cGy. Dose differences showed weak but significant correlations with all geometric metrics (R2 between 0.030 and 0.314).ConclusionsThe autosegmentation methods offering the best combination of performance and execution time are DL and IPP_1. Dose reconstruction on on‐board T2‐weighted MRIs is feasible with autosegmented structures with minimal dosimetric variation from ground truth, but contours should be visually inspected prior to dose reconstruction in an end‐to‐end dose accumulation workflow.