Image-Based Deep Neural Network for Individualizing Radiotherapy Dose Is Transportable Across Health Systems.

PURPOSE: We developed a deep neural network that queries the lung computed tomography-derived feature space to identify radiation sensitivity parameters that can predict treatment failures and hence guide the individualization of radiotherapy dose. In this article, we examine the transportability of this model across health systems. METHODS: This multicenter cohort-based registry included 1,120 patients with cancer in the lung treated with stereotactic body radiotherapy. Pretherapy lung computed tomography images from the internal study cohort (n = 849) were input into a multitask deep neural network to generate an image fingerprint score that predicts time to local failure. Deep learning (DL) scores were input into a regression model to derive iGray, an individualized radiation dose estimate that projects a treatment failure probability of < 5% at 24 months. We validated our findings in an external, holdout cohort (n = 271). RESULTS: There were substantive differences in the baseline patient characteristics of the two study populations, permitting an assessment of model transportability. In the external cohort, radiation treatments in patients with high DL scores failed at a significantly higher rate with 3-year cumulative incidences of local failure of 28.5% (95% CI, 19.8 to 37.8) versus 10.2% (95% CI, 5.9 to 16.2; hazard ratio, 3.3 [95% CI, 1.74 to 6.49]; P < .001). A model that included DL score alone predicted treatment failures with a concordance index of 0.68 (95% CI, 0.59 to 0.77), which had a similar performance to a nested model derived from within the internal cohort (0.70 [0.64 to 0.75]). External cohort patients with iGray values that exceeded the delivered doses had proportionately higher rates of local failure (P < .001). CONCLUSION: Our results support the development and implementation of new DL-guided treatment guidance tools in the image-replete and highly standardized discipline of radiation oncology.

Predicting Treatment Response From Extracellular Volume Fraction for Chemoradiation Therapy of Pancreatic Cancer.

PURPOSE: Dual-energy computed tomography (DECT) data can be used to calculate the extracellular volume fraction (ECVf) in tumors, which has been correlated with treatment outcome. This study sought to find a correlation between ECVf and treatment response as measured by the change in cancer antigen (CA) 19 to 9 during chemoradiation therapy (CRT) for pancreatic cancer. METHODS AND MATERIALS: Dual-energy CT data acquired during the late arterial contrast phase in the standard radiation therapy simulation on a dual-source DECT simulator for 25 patients with pancreatic cancer, along with their CA19-9 and hematocrit data, were analyzed. Each patient underwent preoperative CRT with a prescription of 50.4 Gy in 28 fractions. The patients were chosen based on the presence of a solid tumor in the pancreas that could be clearly delineated. A region of interest (ROI) was placed in the tumor and in the aorta. From the ratio of the iodine density calculated from the DECT in the ROI and the hematocrit taken at the time of simulation, the ECVf was calculated. The ECVf was then compared with the change in CA19-9 before and after the CRT. Distant metastases as the cause of CA19-9 elevation were ruled out on subsequent restaging images before surgery. The DECT-derived iodine ratio was validated using a phantom study. RESULTS: The DECT-derived iodine concentration agreed with the phantom measurements (R2, 1.0). The average hematocrit, ECVf, and change in CA19-9 during the treatment for the 25 patients was 35.6 ± 5.4%, 7.3 ± 4.9%, and -4.6 ± 21.8 respectively. A linear correlation was found between the ECVf and the change in CA19-9, with an R2 of 0.7: ΔCA19-9 = 3.63 × ECVf - 31.1. The correlation was statistically significant (P = .006). CONCLUSIONS: The calculated ECV fraction based on iodine maps from dual-source DECT may be used to predict treatment response after neoadjuvant chemoradiation therapy for pancreatic cancer.

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

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

Deterministic small-scale undulations of image-based risk predictions from the deep learning of lung tumors in motion.

INTRODUCTION: Deep learning (DL) models that use medical images to predict clinical outcomes are poised for clinical translation. For tumors that reside in organs that move, however, the impact of motion (i.e., degenerated object appearance or blur) on DL model accuracy remains unclear. We examine the impact of tumor motion on an image-based DL framework that predicts local failure risk after lung stereotactic body radiotherapy (SBRT). METHODS: We input pre-therapy free breathing (FB) computed tomography (CT) images from 849 patients treated with lung SBRT into a multitask deep neural network to generate an image fingerprint signature (or DL score) that predicts time-to-event local failure outcomes. The network includes a convolutional neural network encoder for extracting imaging features and building a task-specific fingerprint, a decoder for estimating handcrafted radiomic features, and a task-specific network for generating image signature for radiotherapy outcome prediction. The impact of tumor motion on the DL scores was then examined for a holdout set of 468 images from 39 patients comprising: (1) FB CT, (2) four-dimensional (4D) CT, and (3) maximum-intensity projection (MIP) images. Tumor motion was estimated using a 3D vector of the maximum distance traveled, and its association with DL score variance was assessed by linear regression. FINDINGS: The variance and amplitude in 4D CT image-derived DL scores were associated with tumor motion (R2  = 0.48 and 0.46, respectively). Specifically, DL score variance was deterministic and represented by sinusoidal undulations in phase with the respiratory cycle. DL scores, but not tumor volumes, peaked near end-exhalation. The mean of the scores derived from 4D CT images and the score obtained from FB CT images were highly associated (Pearson r = 0.99). MIP-derived DL scores were significantly higher than 4D- or FB-derived risk scores (p < 0.0001). INTERPRETATION: An image-based DL risk score derived from a series of 4D CT images varies in a deterministic, sinusoidal trajectory in a phase with the respiratory cycle. These results indicate that DL models of tumors in motion can be robust to fluctuations in object appearance due to movement and can guide standardization processes in the clinical translation of DL models for patients with lung cancer.

Characterization of size-specific effects during dual-energy CT material decomposition of non-iodine materials.

PURPOSE: The dual-energy CT (DECT) LiverVNC application class in the Siemens Syngo.via software has been used to perform non-iodine material decompositions. However, the LiverVNC application is designed with an optional size-specific calibration based on iodine measurements. This work investigates the effects of this iodine-based size-specific calibration on non-iodine material decomposition and benchmarks alternative methods for size-specific calibrations. METHODS: Calcium quantification was performed with split-filter and sequential-scanning DECT techniques on the Siemens SOMATOM Definition Edge CT scanner. Images were acquired of the Gammex MECT abdomen and head phantom containing calcium inserts with concentrations ranging from 50-300 mgCa/ml. Several workflows were explored investigating the effects of size-specific dual-energy ratios (DERs) and the beam hardening correction (BHC) function in the LiverVNC application. Effects of image noise were also investigated by varying CTDIvol and using iterative reconstruction (ADMIRE). RESULTS: With the default BHC activated, Syngo.via underestimated the calcium concentrations in the abdomen for sequential-scanning acquisitions, leaving residual calcium in the virtual non-contrast images and underestimating calcium in the enhancement images for all DERs. Activation of the BHC with split-filter images resulted in a calcium over- or underestimation depending on the DER. With the BHC inactivated, the use of a single DER led to an under- or overestimate of calcium concentration depending on phantom size and DECT modality. Optimal results were found with BHC inactivated using size-specific DERs. CTDIvol levels and ADMIRE had no significant effect on results. CONCLUSION: When performing non-iodine material decomposition in the LiverVNC application class, it is important to understand the implications of the BHC function and to account for patient size appropriately. The BHC in the LiverVNC application is specific to iodine and leads to inaccurate quantification of other materials. The inaccuracies can be overcome by deactivating the BHC function and using size-specific DERs, which provided the most accurate calcium quantification.

Technical Note: Using virtual noncontrast images from dual-energy CT to eliminate the need of precontrast CT for x-ray radiation treatment planning of abdominal tumors†.

PURPOSE: Radiation therapy (RT) planning frequently utilizes contrast-enhanced CT. However, dose calculations should not be performed on a contrast-enhanced CT because the patient will not receive bolus during treatment. It is typical to acquire CT twice during RT simulation: once before injection of bolus and once after. The registration between these datasets introduces errors. In this work, we investigate the use of virtual noncontrast images (VNC) derived from dual-energy CT (DECT) to eliminate the precontrast CT and the registration error. METHODS: CT datasets, including conventional 120 kVp pre- and postcontrast CTs and postcontrast DECT, acquired for ten pancreatic cancer patients were evaluated. The DECTs were acquired simultaneously using a dual source (DS) CT simulator. VNC and virtual mono-energetic images (VMI) were derived from DECTs. Gross tumor volumes (GTV), planning target volumes (PTV), and organs at risks (OAR) were delineated on the postcontrast CT and then populated to the precontrast CT and the VNC. An IMRT plan (50.4 Gy in 28 fractions) was then optimized on the precontrast CT. Dose distributions were recalculated on the VNC images. Contours from the pre- and postcontrast CTs and the dose distributions based on both were compared. RESULTS: On average, the distance of centroids of the populated duodenum contours on precontrast CT differed by 6.0 ± 4.0 mm from those on postcontrast CTs. The dose distributions on the precontrast CT and VNC were almost identical. The PTV mean and maximum doses differed by 0.1% and 0.2% between the two plans, respectively. CONCLUSION: The VNC derived from DECT can be used to replace the conventional precontrast CT scan for RT planning, eliminating the need for an additional precontrast CT scan and eliminating the registration errors. Thus, VNC can become an important asset to the future of RT.

Characterization of CT Hounsfield Units for 3D acquisition trajectories on a dedicated breast CT system.

Hounsfield Units (HU) are used clinically in differentiating tissue types in a reconstructed CT image, and therefore the HU accuracy of a system is important, especially when using multiple sources, novel detector and non-traditional trajectories. Dedicated clinical breast CT (BCT) systems therefore should be similarly evaluated. In this study, uniform cylindrical phantoms filled with various uniform density fluids were used to characterize differences in HU values between simple circular and complex 3D (saddle) orbits. Based on ACR recommendations, the HU accuracy, center-to-edge variability within a slice, and overall variability within the reconstructed volume were characterized for simple and complex acquisitions possible on a single versatile BCT system. Results illustrate the statistically significantly better performance of the saddle orbit, especially close to the chest and nipple regions of what would clinically be a pendant breast volume. The incomplete cone beam acquisition of a simple circular orbit causes shading artifacts near the nipple, due to insufficient sampling, rendering a major portion of the scanned phantom unusable, whereas the saddle orbit performs exceptionally well and provides a tighter distribution of HU values throughout the reconstructed volumes. This study further establishes the advantages of using 3D acquisition trajectories for breast CT as well as other applications by demonstrating the robustness of HU values throughout large reconstructed volumes.

Contrast-Induced Nephropathy in STEMI Patients With and Without Chronic Kidney Disease.

INTRODUCTION: Contrast-induced nephropathy (CIN) following percutaneous coronary intervention (PCI) is associated with adverse outcomes; however, there are scarce data comparing clinical outcomes of post-PCI CIN in ST elevation myocardial infarction (STEMI) patients with and without chronic kidney disease (CKD). We sought to assess the incidence, clinical predictors, and short-term and long-term clinical outcomes of post-PCI CIN in STEMI patients with and without CKD. METHODS: We performed a retrospective observational cohort study involving 554 patients who underwent PCI for STEMI from February 2010 to November 2013. CKD was defined as estimated glomerular filtration rate ≤60 mL/min and CIN as creatinine increase by ≥25% or ≥0.5 mg/dL from baseline within 72 hours after catheterization contrast exposure. RESULTS: In the entire population, CIN developed in 89 (16%) patients. The incidence of CIN was 19.7% (27/137) in CKD patients and 11.1% (62/417) in non-CKD patients, P < 0.05. Univariate analysis predictors of CIN were older age (65 vs. 60 years), diabetes (35% vs. 21%), peripheral artery disease (11% vs. 5%), cardiogenic shock (24% vs. 13%), hemodynamic support placement (34% vs. 14%), and Mehran score (9.4 ± 7 vs. 5.4 ± 5.2) with all P < 0.05. The predictors of CIN were the same across the CKD and non-CKD cohort with the exception of diabetes. In multivariate analysis, the strongest predictor of CIN in CKD patients was diabetes (odds ratio, 5.8; CI, 1.8-18.6); however, diabetes was not a predictor in the non-CKD population. In the non-CKD population, each single unit increase in the Mehran score was associated with a 1.1 times greater likelihood of CIN (odds ratio, 1.1; CI, 1.01-1.2). Patients with CIN had higher rates of inpatient mortality (14.6% vs. 2.8%), longer length of hospitalization (8 ± 11 vs. 3.4 ± 4.4 days), need for inpatient dialysis (11.2% vs. 0%), higher 30-day mortality (14.6% vs. 3.0%), and higher incidence of long-term serum creatinine >0.5 mg/dL from baseline (16.9% vs. 2.4%) with all P < 0.05. CONCLUSIONS: Overall, we found that CKD patients undergoing PCI for STEMI have a higher incidence of CIN than non-CKD patients. CIN confers worse short-term and long-term outcomes irrespective of baseline renal function.

Implementation and CT sampling characterization of a third-generation SPECT-CT system for dedicated breast imaging.

Stand-alone cone beam computed tomography (CT) and single-photon emission computed tomography (SPECT) systems capable of complex acquisition trajectories have previously been developed for breast imaging. Fully three-dimensional (3-D) motions of SPECT systems provide views into the chest wall and throughout the entire volume. The polar tilting capability of the CBCT system has shown improvement in sampling close to the chest wall, while eliminating cone beam artifacts. Here, a single hybrid SPECT-CT system, with each individual modality capable of independently traversing complex trajectories around a common pendant breast volume, was developed. We present the practical implementation of this design and preliminary results of the CT system. The fully 3-D SPECT was nested inside the suspended CT gantry and oriented perpendicular to the CT source-detector pair. Both subsystems were positioned on a rotation stage, with the combined polar and azimuthal motions enabling spherical trajectories. Six trajectories were used for initial evaluation of the tilt capable CT system. The developed system can achieve polar tilt angles with a [Formula: see text] positioning error and no hysteresis. Initial imaging results demonstrate that additional off-axis projection views of various geometric resolution phantoms facilitate more complete sampling, more consistent attenuation value recovery, and markedly improved reconstructions. This system could have various applications in diagnostic or therapeutic breast imaging.

Deglutition syncope.

Deglutition syncope, also known as swallow syncope, is a neurally mediated reflex syndrome. The common intervention of the heart, esophagus, and stomach by the vagus nerve is central to its pathogenesis, whereby swallowing causes inhibition of the cardiac conduction system. It is most commonly associated with disorders of the esophagus, both organic and functional. Herein we describe the case of a 48-year-old man presenting with transient syncopal episodes that occurred while eating caused by an intrathoracic stomach due to a hiatal hernia.