MU variability in CBCT-guided online adaptive radiation therapy.

PURPOSE: CBCT-guided online-adaptive radiotherapy (oART) systems have been made possible by using artificial intelligence and automation to substantially reduce treatment planning time during on-couch adaptive sessions. Evaluating plans generated during an adaptive session presents significant challenges to the clinical team as the planning process gets compressed into a shorter window than offline planning. We identified MU variations up to 30% difference between the adaptive plan and the reference plan in several oART sessions that caused the clinical team to question the accuracy of the oART dose calculation. We investigated the cause of MU variation and the overall accuracy of the dose delivered when MU variations appear unnecessarily large. METHODS: Dosimetric and adaptive plan data from 604 adaptive sessions of 19 patients undergoing CBCT-guided oART were collected. The analysis included total MU per fraction, planning target volume (PTV) and organs at risk (OAR) volumes, changes in PTV-OAR overlap, and DVH curves. Sessions with MU greater than two standard deviations from the mean were reoptimized offline, verified by an independent calculation system, and measured using a detector array. RESULTS: MU variations relative to the reference plan were normally distributed with a mean of -1.0% and a standard deviation of 11.0%. No significant correlation was found between MU variation and anatomic changes. Offline reoptimization did not reliably reproduce either reference or on-couch total MUs, suggesting that stochastic effects within the oART optimizer are likely causing the variations. Independent dose calculation and detector array measurements resulted in acceptable agreement with the planned dose. CONCLUSIONS: MU variations observed between oART plans were not caused by any errors within the oART workflow. Providers should refrain from using MU variability as a way to express their confidence in the treatment planning accuracy. Clinical decisions during on-couch adaptive sessions should rely on validated secondary dose calculations to ensure optimal plan selection.

Prostate segmentation accuracy using synthetic MRI for high-dose-rate prostate brachytherapy treatment planning.

Objective. Both computed tomography (CT) and magnetic resonance imaging (MRI) images are acquired for high-dose-rate (HDR) prostate brachytherapy patients at our institution. CT is used to identify catheters and MRI is used to segment the prostate. To address scenarios of limited MRI access, we developed a novel generative adversarial network (GAN) to generate synthetic MRI (sMRI) from CT with sufficient soft-tissue contrast to provide accurate prostate segmentation without MRI (rMRI).Approach. Our hybrid GAN, PxCGAN, was trained utilizing 58 paired CT-MRI datasets from our HDR prostate patients. Using 20 independent CT-MRI datasets, the image quality of sMRI was tested using mean absolute error (MAE), mean squared error (MSE), peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM). These metrics were compared with the metrics of sMRI generated using Pix2Pix and CycleGAN. The accuracy of prostate segmentation on sMRI was evaluated using the Dice similarity coefficient (DSC), Hausdorff distance (HD) and mean surface distance (MSD) on the prostate delineated by three radiation oncologists (ROs) on sMRI versus rMRI. To estimate inter-observer variability (IOV), these metrics between prostate contours delineated by each RO on rMRI and the prostate delineated by treating RO on rMRI (gold standard) were calculated.Main results. Qualitatively, sMRI images show enhanced soft-tissue contrast at the prostate boundary compared with CT scans. For MAE and MSE, PxCGAN and CycleGAN have similar results, while the MAE of PxCGAN is smaller than that of Pix2Pix. PSNR and SSIM of PxCGAN are significantly higher than Pix2Pix and CycleGAN (p < 0.01). The DSC for sMRI versus rMRI is within the range of the IOV, while the HD for sMRI versus rMRI is smaller than the HD for the IOV for all ROs (p ≤ 0.03).Significance. PxCGAN generates sMRI images from treatment-planning CT scans that depict enhanced soft-tissue contrast at the prostate boundary. The accuracy of prostate segmentation on sMRI compared to rMRI is within the segmentation variation on rMRI between different ROs.

Dosimetric and workflow impact of synthetic-MRI use in prostate high-dose-rate brachytherapy.

PURPOSE: Target and organ delineation during prostate high-dose-rate (HDR) brachytherapy treatment planning can be improved by acquiring both a postimplant CT and MRI. However, this leads to a longer treatment delivery workflow and may introduce uncertainties due to anatomical motion between scans. We investigated the dosimetric and workflow impact of MRI synthesized from CT for prostate HDR brachytherapy. METHODS AND MATERIALS: Seventy-eight CT and T2-weighted MRI datasets from patients treated with prostate HDR brachytherapy at our institution were retrospectively collected to train and validate our deep-learning-based image-synthesis method. Synthetic MRI was assessed against real MRI using the dice similarity coefficient (DSC) between prostate contours drawn using both image sets. The DSC between the same observer's synthetic and real MRI prostate contours was compared with the DSC between two different observers' real MRI prostate contours. New treatment plans were generated targeting the synthetic MRI-defined prostate and compared with the clinically delivered plans using target coverage and dose to critical organs. RESULTS: Variability between the same observer's prostate contours from synthetic and real MRI was not significantly different from the variability between different observer's prostate contours on real MRI. Synthetic MRI-planned target coverage was not significantly different from that of the clinically delivered plans. There were no increases above organ institutional dose constraints in the synthetic MRI plans. CONCLUSIONS: We developed and validated a method for synthesizing MRI from CT for prostate HDR brachytherapy treatment planning. Synthetic MRI use may lead to a workflow advantage and removal of CT-to-MRI registration uncertainty without loss of information needed for target delineation and treatment planning.

The Aneurysm Occlusion Assistant, an AI platform for real time surgical guidance of intracranial aneurysms.

PURPOSE: In recent years, endovascular treatment has become the dominant approach to treat intracranial aneurysms (IAs). Despite tremendous improvement in surgical devices and techniques, 10-30% of these surgeries require retreatment. Previously, we developed a method which combines quantitative angiography with data-driven modeling to predict aneurysm occlusion within a fraction of a second. This is the first report on a semi-autonomous system, which can predict the surgical outcome of an IA immediately following device placement, allowing for therapy adjustment. Additionally, we previously reported various algorithms which can segment IAs, extract hemodynamic parameters via angiographic parametric imaging, and perform occlusion predictions. METHODS: We integrated these features into an Aneurysm Occlusion Assistant (AnOA) utilizing the Kivy library's graphical instructions and unique language properties for interface development, while the machine learning algorithms were entirely developed within Keras, Tensorflow and skLearn. The interface requires pre- and post-device placement angiographic data. The next steps for aneurysm segmentation, angiographic analysis and prediction have been integrated allowing either autonomous or interactive use. RESULTS: The interface allows for segmentation of IAs and cranial vasculature with a dice index of ~0.78 and prediction of aneurysm occlusion at six months with an accuracy 0.84, in 6.88 seconds. CONCLUSION: This is the first report on the AnOA to guide endovascular treatment of IAs. While this initial report is on a stand-alone platform, the software can be integrated in the angiographic suite allowing direct communication with the angiographic system for a completely autonomous surgical guidance solution.

Investigation of the efficacy of a data-driven CT artifact correction scheme for sparse and truncated projection data for intracranial hemorrhage diagnosis.

Data-driven CT-image reconstruction techniques for truncated or sparsely acquired projection data to reduce radiation dose, iodine volume, and patient motion artifacts have been proposed. These approaches have shown good performance and preservation of image quality metrics. To continue these efforts, we investigated whether these techniques affect the performance of a machine-learning algorithm to identify the presence of intracranial hemorrhage (ICH). Ten-thousand head CT scans were collected from the 2019 RSNA Intracranial Hemorrhage Detection and Classification Challenge dataset. Sinograms were simulated and then resampled in both a one-third truncated and one-third sparse manner. GANs were tasked with correcting the incomplete projection data in two ways. Firstly, in the sinogram domain, where the incomplete sinogram was filled by the GAN and then reconstructed. Secondly, in the reconstruction domain, where the incomplete data were first reconstructed and the sparse or truncation artifacts were corrected by the GAN. Eighty-five hundred images were used for artifact correction network training, and 1500 were withheld for network assessment via an already trained machine-learning algorithm tasked with diagnosis of ICH presence. Fully-sampled reconstructions were compared with the sparse and truncated reconstructions for classification accuracy. Difference in classification accuracy between the fully sampled (83.4%), sparse (82.0%), and truncated (82.3%) reconstructions was minimal, demonstrating that the network diagnosis performance is unaffected by 2/3 reduction of projection data. This work indicates that data-driven reconstructions for a sparse or truncated projection dataset can provide high diagnostic performance for ICH detection at a fraction of the typical radiation dose.

Use of a convolutional neural network to identify infarct core using computed tomography perfusion parameters.

PURPOSE: Computed tomography perfusion (CTP) is used to diagnose ischemic strokes through contralateral hemisphere comparisons of various perfusion parameters. Various perfusion parameter thresholds have been utilized to segment infarct tissue due to differences in CTP software and patient baseline hemodynamics. This study utilized a convolutional neural network (CNN) to eliminate the need for non-universal parameter thresholds to segment infarct tissue. METHODS: CTP data from 63 ischemic stroke patients was retrospectively collected and perfusion parameter maps were generated using Vitrea CTP software. Infarct ground truth labels were segmented from diffusion-weighted imaging (DWI) and CTP and DWI volumes were registered. A U-net based CNN was trained and tested five separate times using each CTP parameter (cerebral blood flow (CBF), cerebral blood volume (CBV), time-to-peak (TTP), mean-transit-time (MTT), delay time). 8,352 infarct slices were utilized with a 60:30:10 training:testing:validation split and Monte Carlo cross-validation was conducted using 20 iterations. Infarct volumes were reconstructed following segmentation from each CTP slice. Infarct spatial and volumetric agreement was compared between each CTP parameter and DWI. RESULTS: Spatial agreement metrics (Dice coefficient, positive predictive value) for each CTP parameter in predicting infarct volumes are: CBF=(0.67, 0.76), CBV=(0.44, 0.62), TTP=(0.60, 0.67), MTT=(0.58, 0.62), delay time=(0.57, 0.60). 95% confidence intervals for volume differences with DWI infarct are: CBF=14.3±11.5 mL, CBV=29.6±21.2 mL, TTP=7.7±15.2 mL, MTT=-10.7±18.6 mL, delay time=-5.7±23.6 mL. CONCLUSIONS: CBF is the most accurate CTP parameter in segmenting infarct tissue. Segmentation of infarct using a CNN has the potential to eliminate non-universal CTP contralateral hemisphere comparison thresholds.

Use of biplane quantitative angiographic imaging with ensemble neural networks to assess reperfusion status during mechanical thrombectomy.

Digital subtraction angiography (DSA) is the main imaging modality used to assess reperfusion during mechanical thrombectomy (MT) when treating large vessel occlusion (LVO) ischemic strokes. To improve this visual and subjective assessment, hybrid models combining angiographic parametric imaging (API) with deep learning tools have been proposed. These models use convolutional neural networks (CNN) with single view individual API maps, thus restricting use of complementary information from multiple views and maps resulting in loss of relevant clinical information. This study investigates use of ensemble networks to combine hemodynamic information from multiple bi-plane API maps to assess level of reperfusion. Three-hundred-eighty-three anteroposterior (AP) and lateral view DSAs were retrospectively collected from patients who underwent MTs of anterior circulation LVOs. API peak height (PH) and area under time density curve (AUC) maps were generated. CNNs were developed to classify maps as adequate/inadequate reperfusion as labeled by two neuro-interventionalists. Outputs from individual networks were combined by weighting each output, using a grid search algorithm. Ensembled, AP-AUC, AP-PH, lateral-AUC, and lateral-PH networks achieved accuracies of 83.0% (95% confidence-interval: 81.2%-84.8%), 74.4% (72.0%-76.7%), 74.2% (72.8%-75.7%), 74.9% (72.2%-77.7%), and 76.9% (74.4%-79.5%); area under receiver operating characteristic curves of 0.86 (0.84-0.88), 0.81 (0.79-0.83), 0.83 (0.81-0.84), 0.82 (0.8-0.84), and 0.84 (0.82-0.87); and Matthews correlation coefficients of 0.66 (0.63-0.70), 0.48 (0.43-0.53), 0.49 (0.46-0.52), 0.51 (0.45-0.56), and 0.54 (0.49-0.59) respectively. Ensembled network performance was significantly better than individual networks (McNemar's p-value<0.05). This study proved feasibility of using ensemble networks to combine hemodynamic information from multiple bi-plane API maps to assess level of reperfusion during MTs.

Investigation of convolutional neural networks using multiple computed tomography perfusion maps to identify infarct core in acute ischemic stroke patients.

Purpose: To assess acute ischemic stroke (AIS) severity, infarct is segmented using computed tomography perfusion (CTP) software, such as RAPID, Sphere, and Vitrea, relying on contralateral hemisphere thresholds. Since this approach is potentially patient dependent, we investigated whether convolutional neural networks (CNNs) could achieve better performances without the need for contralateral hemisphere thresholds. Approach: CTP and diffusion-weighted imaging (DWI) data were retrospectively collected for 63 AIS patients. Cerebral blood flow (CBF), cerebral blood volume (CBV), time-to-peak, mean-transit-time (MTT), and delay time maps were generated using Vitrea CTP software. U-net shaped CNNs were developed, trained, and tested for 26 different input CTP parameter combinations. Infarct labels were segmented from DWI volumes registered with CTP volumes. Infarct volumes were reconstructed from two-dimensional CTP infarct segmentations. To remove erroneous segmentations, conditional random field (CRF) postprocessing was applied and compared with prior results. Spatial and volumetric infarct agreement was assessed between DWI and CTP (CNNs and commercial software) using median infarct difference, median absolute error, dice coefficient, positive predictive value. Results: The most accurate combination of parameters for CNN segmenting infarct using CRF postprocessing was CBF, CBV, and MTT (4.83 mL, 10.14 mL, 0.66, 0.73). Commercial software results are: RAPID = (2.25 mL, 21.48 mL, 0.63, 0.70), Sphere = (7.57 mL, 17.74 mL, 0.64, 0.70), Vitrea = (6.79 mL, 15.28 mL, 0.63, 0.72). Conclusions: Use of CNNs with multiple input perfusion parameters has shown to be accurate in segmenting infarcts and has the ability to improve clinical workflow by eliminating the need for contralateral hemisphere comparisons.

Use of quantitative angiographic methods with a data-driven model to evaluate reperfusion status (mTICI) during thrombectomy.

PURPOSE: Intra-procedural assessment of reperfusion during mechanical thrombectomy (MT) for emergent large vessel occlusion (LVO) stroke is traditionally based on subjective evaluation of digital subtraction angiography (DSA). However, semi-quantitative diagnostic tools which encode hemodynamic properties in DSAs, such as angiographic parametric imaging (API), exist and may be used for evaluation of reperfusion during MT. The objective of this study was to use data-driven approaches, such as convolutional neural networks (CNNs) with API maps, to automatically assess reperfusion in the neuro-vasculature during MT procedures based on the modified thrombolysis in cerebral infarction (mTICI) scale. METHODS: DSAs from patients undergoing MTs of anterior circulation LVOs were collected, temporally cropped to isolate late arterial and capillary phases, and quantified using API peak height (PH) maps. PH maps were normalized to reduce injection variability. A CNN was developed, trained, and tested to classify PH maps into 2 outcomes (mTICI 0,1,2a/mTICI 2b,2c,3) or 3 outcomes (mTICI 0,1,2a/mTICI 2b/mTICI 2c,3), respectively. Ensembled networks were used to combine information from multiple views (anteroposterior and lateral). RESULTS: The study included 383 DSAs. For the 2-outcome classification, average accuracy was 81.0% (95% CI, 79.0-82.9%), and the area under the receiver operating characteristic curve (AUROC) was 0.86 (0.84-0.88). For the 3-outcome classification, average accuracy was 64.0% (62.0-66.0), and AUROC values were 0.85 (0.83-0.87), 0.74 (0.71-0.77), and 0.78 (0.76-0.81) for the mTICI 0,1,2a, mTICI 2b, and mTICI 2c,3 classes, respectively. CONCLUSION: This study demonstrated the feasibility of using hemodynamic information in API maps with data-driven models to autonomously assess intra-procedural reperfusion during MT.

Enhancing performance of a computed tomography perfusion software for improved prediction of final infarct volume in acute ischemic stroke patients.

Computed tomography perfusion (CTP) is crucial for acute ischemic stroke (AIS) patient diagnosis. To improve infarct prediction, enhanced image processing and automated parameter selection have been implemented in Vital Images' new CTP+ software. We compared CTP+ with its previous version, commercially available software (RAPID and Sphere), and follow-up diffusion-weighted imaging (DWI). Data from 191 AIS patients between March 2019 and January 2020 was retrospectively collected and allocated into endovascular intervention (n = 81) and conservative treatment (n = 110) cohorts. Intervention patients were treated for large vessel occlusion, underwent mechanical thrombectomy, and achieved successful reperfusion of thrombolysis in cerebral infarction 2b/2c/3. Conservative treatment patients suffered large or small vessel occlusion and did not receive intravenous thrombolysis or mechanical thrombectomy. Infarct and penumbra were assessed using intervention and conservative treatment patients, respectively. Infarct and penumbra volumes were segmented from CTP+ and compared with 24-h DWI along with RAPID, Sphere, and Vitrea. Mean infarct differences (95% confidence intervals) and Spearman correlation coefficients (SCCs) between DWI and each CTP software product for intervention patients are: CTP+ = (5.8 ± 5.9 ml, 0.62), RAPID = (10.0 ± 5.2 ml, 0.73), Sphere = (3.0 ± 6.0 ml, 0.56), Vitrea = (7.2 ± 4.9 ml, 0.66). For conservative treatment patients, mean infarct differences and SCCs are: CTP+ = (-8.0 ± 5.4 ml, 0.64), RAPID = (-25.6 ± 11.5 ml, 0.60), Sphere = (-25.6 ± 8.0 ml, 0.66), Vitrea = (1.3 ± 4.0 ml, 0.72). CTP+ performed similarly to RAPID and Sphere in addition to its semi-automated predecessor, Vitrea, when assessing intervention patient infarct volumes. For conservative treatment patients, CTP+ outperformed RAPID and Sphere in assessing penumbra. Semi-automated Vitrea remains the most accurate in assessing penumbra, but CTP+ provides an improved workflow from its predecessor.

Assessment of computed tomography perfusion software in predicting spatial location and volume of infarct in acute ischemic stroke patients: a comparison of Sphere, Vitrea, and RAPID.

BACKGROUND: CT perfusion (CTP) infarct and penumbra estimations determine the eligibility of patients with acute ischemic stroke (AIS) for endovascular intervention. This study aimed to determine volumetric and spatial agreement of predicted RAPID, Vitrea, and Sphere CTP infarct with follow-up fluid attenuation inversion recovery (FLAIR) MRI infarct. METHODS: 108 consecutive patients with AIS and large vessel occlusion were included in the study between April 2019 and January 2020 . Patients were divided into two groups: endovascular intervention (n=58) and conservative treatment (n=50). Intervention patients were treated with mechanical thrombectomy and achieved successful reperfusion (Thrombolysis in Cerebral Infarction 2b/2 c/3) while patients in the conservative treatment group did not receive mechanical thrombectomy or intravenous thrombolysis. Intervention and conservative treatment patients were included to assess infarct and penumbra estimations, respectively. It was assumed that in all patients treated conservatively, penumbra converted to infarct. CTP infarct and penumbra volumes were segmented from RAPID, Vitrea, and Sphere to assess volumetric and spatial agreement with follow-up FLAIR MRI. RESULTS: Mean infarct differences (95% CIs) between each CTP software and FLAIR MRI for each cohort were: intervention cohort: RAPID=9.0±7.7 mL, Sphere=-0.2±8.7 mL, Vitrea=-7.9±8.9 mL; conservative treatment cohort: RAPID=-31.9±21.6 mL, Sphere=-26.8±17.4 mL, Vitrea=-15.3±13.7 mL. Overlap and Dice coefficients for predicted infarct were (overlap, Dice): intervention cohort: RAPID=(0.57, 0.44), Sphere=(0.68, 0.60), Vitrea=(0.70, 0.60); conservative treatment cohort: RAPID=(0.71, 0.56), Sphere=(0.73, 0.60), Vitrea=(0.72, 0.64). CONCLUSIONS: Sphere proved the most accurate in patients who had intervention infarct assessment as Vitrea and RAPID overestimated and underestimated infarct, respectively. Vitrea proved the most accurate in penumbra assessment for patients treated conservatively although all software overestimated penumbra.

CT artifact correction for sparse and truncated projection data using generative adversarial networks.

PURPOSE: Computed tomography image reconstruction using truncated or sparsely acquired projection data to reduce radiation dose, iodine volume, and patient motion artifacts has been widely investigated. To continue these efforts, we investigated the use of machine learning-based reconstruction techniques using deep convolutional generative adversarial networks (DCGANs) and evaluated its effect using standard imaging metrics. METHODS: Ten thousand head computed tomography (CT) scans were collected from the 2019 RSNA Intracranial Hemorrhage Detection and Classification Challenge dataset. Sinograms were simulated and then resampled in both a one-third truncated and one-third sparse manner. DCGANs were tasked with correcting the incomplete projection data, either in the sinogram domain where the full sinogram was recovered by the DCGAN and then reconstructed, or the reconstruction domain where the incomplete data were first reconstructed and the sparse or truncation artifacts were corrected by the DCGAN. Seventy-five hundred images were used for network training and 2500 were withheld for network assessment using mean absolute error (MAE), structural similarity index measure (SSIM), and peak signal-to-noise ratio (PSNR) between results of different correction techniques. Image data from a quality-assurance phantom were also resampled in the two manners and corrected and reconstructed for network performance assessment using line profiles across high-contrast features, the modulation transfer function (MTF), noise power spectrum (NPS), and Hounsfield Unit (HU) linearity analysis. RESULTS: Better agreement with the fully sampled reconstructions were achieved from sparse acquisition corrected in the sinogram domain and the truncated acquisition corrected in the reconstruction domain. MAE, SSIM, and PSNR showed quantitative improvement from the DCGAN correction techniques. HU linearity of the reconstructions was maintained by the correction techniques for the sparse and truncated acquisitions. MTF curves reached the 10% modulation cutoff frequency at 5.86 lp/cm for the truncated corrected reconstruction compared with 2.98 lp/cm for the truncated uncorrected reconstruction, and 5.36 lp/cm for the sparse corrected reconstruction compared with around 2.91 lp/cm for the sparse uncorrected reconstruction. NPS analyses yielded better agreement across a range of frequencies between the resampled corrected phantom and truth reconstructions. CONCLUSIONS: We demonstrated the use of DCGANs for CT-image correction from sparse and truncated simulated projection data, while preserving imaging quality of the fully sampled projection data.

Effect of computed tomography perfusion post-processing algorithms on optimal threshold selection for final infarct volume prediction.

In acute ischemic stroke (AIS) patients, eligibility for endovascular intervention is commonly determined through computed tomography perfusion (CTP) analysis by quantifying ischemic tissue using perfusion parameter thresholds. However, thresholds are not uniform across all analysis methods due to dependencies on patient demographics and computational algorithms. This study aimed to investigate optimal perfusion thresholds for quantifying infarct and penumbra volumes using two post-processing CTP algorithms: Vitrea Bayesian and singular value decomposition plus (SVD+). We utilized 107 AIS patients (67 non-intervention patients and 40 successful reperfusion of thrombolysis in cerebral infarction (2b/3) patients). Infarct volumes were predicted for both post-processing algorithms through contralateral hemisphere comparisons using absolute time-to-peak (TTP) and relative regional cerebral blood volume (rCBV) thresholds ranging from +2.8 seconds to +9.3 seconds and -0.23 to -0.56 respectively. Optimal thresholds were determined by minimizing differences between predicted CTP and 24-hour fluid-attenuation inversion recovery magnetic resonance imaging infarct. Optimal thresholds were tested on 60 validation patients (30 intervention and 30 non-intervention) and compared using RAPID CTP software. Among the 67 non-intervention and 40 intervention patients, the following optimal thresholds were determined: intervention Bayesian: TTP = +4.8 seconds, rCBV = -0.29; intervention SVD+: TTP = +5.8 seconds, rCBV = -0.29; non-intervention Bayesian: TTP = +5.3 seconds, rCBV = -0.32; non-intervention SVD+: TTP = +6.3 seconds, rCBV = -0.26. When comparing SVD+ and Bayesian post-processing algorithms, optimal thresholds for TTP were significantly different for intervention and non-intervention patients. rCBV optimal thresholds were equal for intervention patients and significantly different for non-intervention patients. Comparison with commercially utilized software indicated similar performance.

Initial evaluation of a convolutional neural network used for noninvasive assessment of coronary artery disease severity from coronary computed tomography angiography data.

PURPOSE: Coronary computed tomography angiography (CTA) has one of the highest diagnostic sensitivities for detection of the significance of coronary artery disease (CAD); however, sensitivity is moderate and may result in increased catheterization rates. We performed an efficacy study to determine whether a trained machine learning algorithm that uses coronary CTA data may improve CAD diagnosis accuracy. METHODS: Sixty-four-patient image datasets based on coronary CTA were retrospectively collected to generate eight views considering 45° increments around the coronary artery centerline. The dataset was randomly split into training and testing cohorts. Invasive FFR measurements were used as ground truth labels. A convolutional neural network (CNN) was trained and the model's capacity to predict severity of CAD was assessed on the testing cohort. Classification accuracy and area under the receiver operating characteristic curve (AUROC) analysis were performed. Similar CAD severity classification accuracy and AUROC analyses were performed using only percent diameter stenosis (%DS) and CT-derived FFR performed by 13 operators with various levels of expertise. RESULTS: Classification accuracy over the test cohort was 80.9% using the trained network and 72.4% using the user-operated CT-derived FFR software. AUROC over the test cohort was 0.862 using the trained network, 0.807 using %DS, and 0.758 using the human-operated CT-derived FFR software. CONCLUSIONS: A trained neural network compared noninferiorly in-terms of classification accuracy and AUROC with human operators of a CT-derived FFR software, and in-terms of AUROC with clinical decision-making using %DS.

Semi-supervised planning method for breast electronic tissue compensation treatments based on breast radius and separation.

BACKGROUND: The aim of the study was to develop and assess a technique for the optimization of breast electronic tissue compensation (ECOMP) treatment plans based on the breast radius and separation. MATERIALS AND METHODS: Ten ECOMP plans for 10 breast cancer patients delivered at our institute were collected for this work. Pre-treatment CT-simulation images were anonymized and input to a framework for estimation of the breast radius and separation for each axial slice. Optimal treatment fluence was estimated based on the breast radius and separation, and a total beam fluence map for both medial and lateral fields was generated. These maps were then imported into the Eclipse Treatment Planning System and used to calculate a dose distribution. The distribution was compared to the original treatment hand-optimized by a medical dosimetrist. An additional comparison was performed by generating plans assuming a single tissue penetration depth determined by averaging the breast radius and separation over the entire treatment volume. Comparisons between treatment plans used the dose homogeneity index (HI; lower number is better). RESULTS: HI was non-inferior between our algorithm (HI = 12.6) and the dosimetrist plans (HI = 9.9) (p-value > 0.05), and was superior than plans obtained using a single penetration depth (HI = 17.0) (p-value < 0.05) averaged over the 10 collected plans. Our semi-supervised algorithm takes approximately 20 seconds for treatment plan generation and runs with minimal user input, which compares favorably with the dosimetrist plans that can take up to 30 minutes of attention for full optimization. CONCLUSIONS: This work indicates the potential clinical utility of a technique for the optimization of ECOMP breast treatments.

Automatic radiomic feature extraction using deep learning for angiographic parametric imaging of intracranial aneurysms.

BACKGROUND: Angiographic parametric imaging (API) is an imaging method that uses digital subtraction angiography (DSA) to characterize contrast media dynamics throughout the vasculature. This requires manual placement of a region of interest over a lesion (eg, an aneurysm sac) by an operator. OBJECTIVE: The purpose of our work was to determine if a convolutional neural network (CNN) was able to identify and segment the intracranial aneurysm (IA) sac in a DSA and extract API radiomic features with minimal errors compared with human user results. METHODS: Three hundred and fifty angiographic images of IAs were retrospectively collected. The IAs and surrounding vasculature were manually contoured and the masks put to a CNN tasked with semantic segmentation. The CNN segmentations were assessed for accuracy using the Dice similarity coefficient (DSC) and Jaccard index (JI). Area under the receiver operating characteristic curve (AUROC) was computed. API features based on the CNN segmentation were compared with the human user results. RESULTS: The mean JI was 0.823 (95% CI 0.783 to 0.863) for the IA and 0.737 (95% CI 0.682 to 0.792) for the vasculature. The mean DSC was 0.903 (95% CI 0.867 to 0.937) for the IA and 0.849 (95% CI 0.811 to 0.887) for the vasculature. The mean AUROC was 0.791 (95% CI 0.740 to 0.817) for the IA and 0.715 (95% CI 0.678 to 0.733) for the vasculature. All five API features measured inside the predicted masks were within 18% of those measured inside manually contoured masks. CONCLUSIONS: CNN segmentation of IAs and surrounding vasculature from DSA images is non-inferior to manual contours of aneurysms and can be used in parametric imaging procedures.

Feasibility study for use of angiographic parametric imaging and deep neural networks for intracranial aneurysm occlusion prediction.

BACKGROUND: Angiographic parametric imaging (API), based on digital subtraction angiography (DSA), is a quantitative imaging tool that may be used to extract contrast flow parameters related to hemodynamic conditions in abnormal pathologies such as intracranial aneurysms (IAs). OBJECTIVE: To investigate the feasibility of using deep neural networks (DNNs) and API to predict IA occlusion using pre- and post-intervention DSAs. METHODS: We analyzed DSA images of IAs pre- and post-treatment to extract API parameters in the IA dome and the corresponding main artery (un-normalized data). We implemented a two-step correction to account for injection variability (normalized data) and projection foreshortening (relative data). A DNN was trained to predict a binary IA occlusion outcome: occluded/unoccluded. Network performance was assessed with area under the receiver operating characteristic curve (AUROC) and classification accuracy. To evaluate the effect of the proposed corrections, prediction accuracy analysis was performed after each normalization step. RESULTS: The study included 190 IAs. The mean and median duration between treatment and follow-up was 9.8 and 8.0 months, respectively. For the un-normalized, normalized, and relative subgroups, the DNN average prediction accuracies for IA occlusion were 62.5% (95% CI 60.5% to 64.4%), 70.8% (95% CI 68.2% to 73.4%), and 77.9% (95% CI 76.2% to 79.6%). The average AUROCs for the same subgroups were 0.48 (0.44-0.52), 0.67 (0.61-0.73), and 0.77 (0.74-0.80). CONCLUSIONS: The study demonstrated the feasibility of using API and DNNs to predict IA occlusion using only pre- and post-intervention angiographic information.

Use of a CMOS-based micro-CT system to validate a ring artifact correction algorithm on low-dose image data.

The imaging of objects using high-resolution detectors coupled to CT systems may be made challenging due to the presence of ring artifacts in the reconstructed data. Not only are the artifacts qualitatilvely distracting, they reduce the SNR of the reconstructed data and may lead to a reduction in the clinical utility of the image data. To address these challenges, we introduce a multistep algorithm that greatly reduces the impact of the ring artifacts on the reconstructed data through image processing in the sinogram space. First, for a single row of detectors corresponding to one slice, we compute the mean of every detector element in the row across all projection view angles and place the reciprocal values in a vector with length equal to the number of detector elements in a row. This vector is then multiplied with each detector element value for each projection view angle, obtaining a normalized or corrected sinogram. This sinogram is subtracted from the original uncorrected sinogram of the slice to obtain a difference map, which is then blurred with a median filter along the row direction. This blurred difference map is summed back to the corrected sinogram, to obtain the final sinogram, which can be back projected to obtain an axial slice of the scanned object, with a greatly reduced presence of ring artifacts. This process is done for each detector row corresponding to each slice. The performance of this algorithm was assessed using images of a mouse femur. These images were acquired using a micro-CT system coupled to a high-resolution CMOS detector. We found that the use of this algorithm led to an increase in SNR and a more uniform line-profile, as a result of the reduction in the presence of the ring artifacts.

Design Optimization for Accurate Flow Simulations in 3D Printed Vascular Phantoms Derived from Computed Tomography Angiography.

3D printing has been used to create complex arterial phantoms to advance device testing and physiological condition evaluation. Stereolithographic (STL) files of patient-specific cardiovascular anatomy are acquired to build cardiac vasculature through advanced mesh-manipulation techniques. Management of distal branches in the arterial tree is important to make such phantoms practicable. We investigated methods to manage the distal arterial flow resistance and pressure thus creating physiologically and geometrically accurate phantoms that can be used for simulations of image-guided interventional procedures with new devices. Patient specific CT data were imported into a Vital Imaging workstation, segmented, and exported as STL files. Using a mesh-manipulation program (Meshmixer) we created flow models of the coronary tree. Distal arteries were connected to a compliance chamber. The phantom was then printed using a Stratasys Connex3 multimaterial printer: the vessel in TangoPlus and the fluid flow simulation chamber in Vero. The model was connected to a programmable pump and pressure sensors measured flow characteristics through the phantoms. Physiological flow simulations for patient-specific vasculature were done for six cardiac models (three different vasculatures comparing two new designs). For the coronary phantom we obtained physiologically relevant waves which oscillated between 80 and 120 mmHg and a flow rate of ~125 ml/min, within the literature reported values. The pressure wave was similar with those acquired in human patients. Thus we demonstrated that 3D printed phantoms can be used not only to reproduce the correct patient anatomy for device testing in image-guided interventions, but also for physiological simulations. This has great potential to advance treatment assessment and diagnosis.

Implementation of material decomposition using an EMCCD and CMOS-based micro-CT system.

This project assessed the effectiveness of using two different detectors to obtain dual-energy (DE) micro-CT data for the carrying out of material decomposition. A micro-CT coupled to either a complementary metal-oxide semiconductor (CMOS) or an electron multiplying CCD (EMCCD) detector was used to acquire image data of a 3D-printed phantom with channels filled with different materials. At any instance, materials such as iohexol contrast agent, water, and platinum were selected to make up the scanned object. DE micro-CT data was acquired, and slices of the scanned object were differentiated by material makeup. The success of the decomposition was assessed quantitatively through the computation of percentage normalized root-mean-square error (%NRMSE). Our results indicate a successful decomposition of iohexol for both detectors (%NRMSE values of 1.8 for EMCCD, 2.4 for CMOS), as well as platinum (%NRMSE value of 4.7). The CMOS detector performed material decomposition on air and water on average with 7 times more %NRMSE, possibly due to the decreased sensitivity of the CMOS system. Material decomposition showed the potential to differentiate between materials such as the iohexol and platinum, perhaps opening the door for its use in the neurovascular anatomical region. Work supported by Toshiba America Medical Systems, and partially supported by NIH grant 2R01EB002873.