Search NASA⌕ Search

SEARCH · Search NASA

Results for “Bootstrap”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 703 records · Page 39

Real-time deep-learning inversion of seismic full waveform data for CO 2 saturation and uncertainty in geological carbon storage monitoring

Deep-learning inversion has recently drawn attention in geological carbon storage research due to its potential of imaging and monitoring carbon storage in real time, significantly improving efficiency and safety of carbon storage operations. We present a deep-learning full waveform inversion method that after the neural network has been trained can image CO 2 saturation and its uncertainty in real time. Our deep-learning inversion method is based on the U-Net architecture with the neural network trained on pairs of synthetic seismic data and CO 2 saturation models. Accordingly, our training establishes a mapping relationship between seismic data and CO 2 saturation models and once fully trained directly estimates CO 2 saturation as a function of subsurface location. We further quantify uncertainties of CO 2 saturation estimates using the Monte Carlo dropout method and a bootstrap aggregating method. For this proof-of-concept study, the CO 2 training models and data are derived from the Kimberlina 1.2 model, a hypothetical 3D geological carbon storage model that is constructed based on various geological and hydrological data from the Southern San Joaquin Basin, California. We perform deep-learning inversion experiments using noise-free and noisy training and test data sets and compare the results. Our modelling experiments show that (1) the deep-learning inversion can estimate 2D distributions of CO 2 fairly well even in the presence of Gaussian random noise and (2) both CO 2 saturation imaging and uncertainty quantification can be done in real time. Our results suggest that the deep-learning inversion method can serve as a robust real-time monitoring tool for geological carbon storage and/or other time-varying reservoir/aquifer properties that result from injection, extraction, and/or other subsurface transport phenomena.

58 GEOSCIENCES↗

Deep learning multiphysics network for imaging CO 2 saturation and estimating uncertainty in geological carbon storage

Multiphysics inversion exploits different types of geophysical data that often complement each other and aims to improve overall imaging resolution and reduce uncertainties in geophysical interpretation. Despite the advantages, traditional multiphysics inversion is challenging because it requires a large amount of computational time and intensive human interactions for preprocessing data and finding trade-off parameters. These issues make it nearly impossible for traditional multiphysics inversion to be applied as a real-time monitoring tool for geological carbon storage. In this paper, we present a deep learning (DL) multiphysics network for imaging CO 2 saturation in real time. The multiphysics network consists of three encoders for analysing seismic, electromagnetic and gravity data and shares one decoder for combining imaging capabilities of the different geophysical data for better predicting CO 2 saturation. The network is trained on pairs of CO 2 label models and multiphysics data so that it can directly image CO 2 saturation. Here we use the bootstrap aggregating method to enhance the imaging accuracy and estimate uncertainties associated with CO 2 saturation images. Using realistic CO 2 label models and multiphysics data derived from the Kimberlina CO 2 storage model, we evaluate the performance of the deep learning multiphysics network and compare its imaging results to those from the deep learning single-physics networks. Our modelling experiments show that the deep learning multiphysics network for seismic, electromagnetic, and gravity data not only improves the imaging accuracy but also reduces uncertainties associated with CO 2 saturation images. Our results also suggest that the deep learning multiphysics network for the non-seismic data (i.e., electromagnetic and gravity) can be used as an effective low-cost monitoring tool in between regular seismic monitoring.

58 GEOSCIENCES↗

A Framework for Inverse Prediction Using Functional Response Data

Inverse prediction models have commonly been developed to handle scalar data from physical experiments. However, it is not uncommon for data to be collected in functional form. When data are collected in functional form, it must be aggregated to fit the form of traditional methods, which often results in a loss of information. For expensive experiments, this loss of information can be costly. In this study, we introduce the functional inverse prediction (FIP) framework, a general approach which uses the full information in functional response data to provide inverse predictions with probabilistic prediction uncertainties obtained with the bootstrap. The FIP framework is a general methodology that can be modified by practitioners to accommodate many different applications and types of data. We demonstrate the framework, highlighting points of flexibility, with a simulation example and applications to weather data and to nuclear forensics. Results show how functional models can improve the accuracy and precision of predictions.

42 ENGINEERING↗

Three-dimensional super line-localization in low signal-to-noise microscope images via prior-apprised unsupervised learning (PAUL)

Biological processes such as processive enzyme turnover and intracellular cargo tracking involve the dynamic motion of a small "article" along a curvilinear biopolymer track. To understand these processes that occur across multiple length and time scales, one must acquire both the trajectory of the particle and the position of the track along which it moves, possibly by combining high-resolution single-particle tracking with conventional microscopy. Yet, usually there is a significant resolution mismatch between these modalities: while the tracked particle is localized with a precision of 10 nm, the image of the surroundings is limited by optical difraction, with 200 nm lateral and 500 nm axial resolutions. Compared to the particle's trajectory, the surrounding curvilinear structure appears as a blurred and noisy image. This disparity in the spatial resolutions of the particle trajectory and the surrounding curvilinear structure image makes data reconstruction, as well as interpretation, particularly challenging. Analysis is further complicated when the curvilinear structures are oriented arbitrarily in 3D space. Here, we present a prior-apprised unsupervised learning (PAUL) approach to extract information from 3D images where the underlying features resemble a curved line such as a filament or microtubule. This three-stage framework starts with a Hessian-based feature enhancement, which is followed by feature registration, where local line segments are detected on repetitively sampled subimage tiles. In the final stage, statistical learning, segments are clustered based on their geometric relationships. Principal curves are then approximated from each segment group via statistical tools including principal component analysis, bootstrap and kernel transformation. This procedure is characterized on simulated images, where sub-voxel medium deviations from true curves have been achieved. The 3D PAUL approach has also been implemented for successful line localization in experimental 3D images of gold nanowires obtained using a multifocal microscope. Lastly, this work not only bridges the resolution gap between two microscopy modalities, but also allows us to conduct 3D super line-localization imaging experiments, without using super-resolution techniques.

42 ENGINEERING↗

Assessing operating characteristics of CAD algorithms in the absence of a gold standard

Purpose: The authors examine potential bias when using a reference reader panel as ''gold standard'' for estimating operating characteristics of CAD algorithms for detecting lesions. As an alternative, the authors propose latent class analysis (LCA), which does not require an external gold standard to evaluate diagnostic accuracy. Methods: A binomial model for multiple reader detections using different diagnostic protocols was constructed, assuming conditional independence of readings given true lesion status. Operating characteristics of all protocols were estimated by maximum likelihood LCA. Reader panel and LCA based estimates were compared using data simulated from the binomial model for a range of operating characteristics. LCA was applied to 36 thin section thoracic computed tomography data sets from the Lung Image Database Consortium (LIDC): Free search markings of four radiologists were compared to markings from four different CAD assisted radiologists. For real data, bootstrap-based resampling methods, which accommodate dependence in reader detections, are proposed to test of hypotheses of differences between detection protocols. Results: In simulation studies, reader panel based sensitivity estimates had an average relative bias (ARB) of -23% to -27%, significantly higher (p-value <0.0001) than LCA (ARB -2% to -6%). Specificity was well estimated by both reader panel (ARB -0.6% to -0.5%) and LCA (ARB 1.4%-0.5%). Among 1145 lesion candidates LIDC considered, LCA estimated sensitivity of reference readers (55%) was significantly lower (p-value 0.006) than CAD assisted readers' (68%). Average false positives per patient for reference readers (0.95) was not significantly lower (p-value 0.28) than CAD assisted readers' (1.27). Conclusions: Whereas a gold standard based on a consensus of readers may substantially bias sensitivity estimates, LCA may be a significantly more accurate and consistent means for evaluating diagnostic accuracy.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Measuring output factors of small fields formed by collimator jaws and multileaf collimator using plastic scintillation detectors

Purpose: As the practice of using high-energy photon beams to create therapeutic radiation fields of subcentimeter dimensions (as in intensity-modulated radiotherapy or stereotactic radiosurgery) grows, so too does the need for accurate verification of beam output at these small fields in which standard practices of dose verification break down. This study investigates small-field output factors measured using a small plastic scintillation detector (PSD), as well as a 0.01 cm{sup 3} ionization chamber. Specifically, output factors were measured with both detectors using small fields that were defined by either the X-Y collimator jaws or the multileaf collimator (MLC). Methods: A PSD of 0.5 mm diameter and 2 mm length was irradiated with 6 and 18 MV linac beams. The PSD was positioned vertically at a source-to-axis distance of 100 cm, at 10 cm depth in a water phantom, and irradiated with fields ranging in size from 0.5x0.5 to 10x10 cm{sup 2}. The field sizes were defined either by the collimator jaws alone or by a MLC alone. The MLC fields were constructed in two ways: with the closed leaves (i.e., those leaves that were not opened to define the square field) meeting at either the field center line or at a 4 cm offset from the center line. Scintillation light was recorded using a CCD camera and an estimation of error in the median-filtered signals was made using the bootstrapping technique. Measurements were made using a CC01 ionization chamber under conditions identical to those used for the PSD. Results: Output factors measured by the PSD showed close agreement with those measured using the ionization chamber for field sizes of 2.0x2.0 cm{sup 2} and above. At smaller field sizes, the PSD obtained output factors as much as 15% higher than those found using the ionization chamber by 0.6x0.6 cm{sup 2} jaw-defined fields. Output factors measured with no offset of the closed MLC leaves were as much as 20% higher than those measured using a 4 cm leaf offset. Conclusions: The authors' results suggest that PSDs provide a useful and possibly superior alternative to existing dosimetry systems for small fields, as they are inherently less susceptible to volume-averaging and perturbation effects than larger, air-filled ionization chambers. Therefore, PSDs may provide more accurate small-field output factor determination, regardless of the collimation mechanism.

61 RADIATION PROTECTION AND DOSIMETRY↗

Computerized method for evaluating diagnostic image quality of calcified plaque images in cardiac CT: Validation on a physical dynamic cardiac phantom

Purpose: In cardiac computed tomography (CT), important clinical indices, such as the coronary calcium score and the percentage of coronary artery stenosis, are often adversely affected by motion artifacts. As a result, the expert observer must decide whether or not to use these indices during image interpretation. Computerized methods potentially can be used to assist in these decisions. In a previous study, an artificial neural network (ANN) regression model provided assessability (image quality) indices of calcified plaque images from the software NCAT phantom that were highly agreeable with those provided by expert observers. The method predicted assessability indices based on computer-extracted features of the plaque. In the current study, the ANN-predicted assessability indices were used to identify calcified plaque images with diagnostic calcium scores (based on mass) from a physical dynamic cardiac phantom. The basic assumption was that better quality images were associated with more accurate calcium scores. Methods: A 64-channel CT scanner was used to obtain 500 calcified plaque images from a physical dynamic cardiac phantom at different heart rates, cardiac phases, and plaque locations. Two expert observers independently provided separate sets of assessability indices for each of these images. Separate sets of ANN-predicted assessability indices tailored to each observer were then generated within the framework of a bootstrap resampling scheme. For each resampling iteration, the absolute calcium score error between the calcium scores of the motion-contaminated plaque image and its corresponding stationary image served as the ground truth in terms of indicating images with diagnostic calcium scores. The performances of the ANN-predicted and observer-assigned indices in identifying images with diagnostic calcium scores were then evaluated using ROC analysis. Results: Assessability indices provided by the first observer and the corresponding ANN performed similarly (AUC{sub OBS1}=0.80 [0.73,0.86] vs AUC{sub ANN1}=0.88 [0.82,0.92]) as that of the second observer and the corresponding ANN (AUC{sub OBS2}=0.87 [0.83,0.91] vs AUC{sub ANN2}=0.90 [0.85,0.94]). Moreover, the ANN-predicted indices were generated in a fraction of the time required to obtain the observer-assigned indices. Conclusions: ANN-predicted assessability indices performed similar to observer-assigned assessability indices in identifying images with diagnostic calcium scores from the physical dynamic cardiac phantom. The results of this study demonstrate the potential of using computerized methods for identifying images with diagnostic clinical indices in cardiac CT images.

60 APPLIED LIFE SCIENCES↗

Patient radiation doses in interventional cardiology in the U.S.: Advisory data sets and possible initial values for U.S. reference levels

Purpose: To determine patient radiation doses from interventional cardiology procedures in the U.S and to suggest possible initial values for U.S. benchmarks for patient radiation dose from selected interventional cardiology procedures [fluoroscopically guided diagnostic cardiac catheterization and percutaneous coronary intervention (PCI)]. Methods: Patient radiation dose metrics were derived from analysis of data from the 2008 to 2009 Nationwide Evaluation of X-ray Trends (NEXT) survey of cardiac catheterization. This analysis used deidentified data and did not require review by an IRB. Data from 171 facilities in 30 states were analyzed. The distributions (percentiles) of radiation dose metrics were determined for diagnostic cardiac catheterizations, PCI, and combined diagnostic and PCI procedures. Confidence intervals for these dose distributions were determined using bootstrap resampling. Results: Percentile distributions (advisory data sets) and possible preliminary U.S. reference levels (based on the 75th percentile of the dose distributions) are provided for cumulative air kerma at the reference point (K{sub a,r}), cumulative air kerma-area product (P{sub KA}), fluoroscopy time, and number of cine runs. Dose distributions are sufficiently detailed to permit dose audits as described in National Council on Radiation Protection and Measurements Report No. 168. Fluoroscopy times are consistent with those observed in European studies, but P{sub KA} is higher in the U.S. Conclusions: Sufficient data exist to suggest possible initial benchmarks for patient radiation dose for certain interventional cardiology procedures in the U.S. Our data suggest that patient radiation dose in these procedures is not optimized in U.S. practice.

07 ISOTOPE AND RADIATION SOURCES↗

WE-E-18A-09: Application of a Channelized Hotelling Observer Model to Evaluate Angiographic Imaging Equipment

Purpose: To develop and apply an observer model to objectively evaluate and compare the performance of different angiographic imaging equipment and acquisition variables. Methods: Image Acquisition— Iodine-based phantoms were created with target diameters: 0.5, 1, 2, and 4 mm. The phantoms were imaged using both planes of a bi-plane angiography system with detector pixel dimensions 0.1542 mm{sup 2} and 0.1842 mm{sup 2}, respectively. All four phantoms were imaged with magnification factors 1.5, 1.25 and 1 and with the large and small focal spots. Phantom position and the dose per frame (0.12– 0.24 μ Gy/frame) were varied for a single phantom size, magnification and focal spot. Observer Model— For each experimental condition, 1200 signal-present and signal-absent images were acquired and a detectability index (d') was calculated with a Gabor-channelized Hotelling observer (CHO) model. Detectability indices were evaluated as a function of dose, phantom size, and magnification. The model was then applied to compare d' of the two imaging planes and focal spots. Uncertainty in d' was estimated by bootstrapping the data and by examining the shift-variance of systems. Results: Detectability indices varied linearly with magnification and the square root of dose. For the 2 and 4 mm phantoms, d' varied linearly with diameter. For the 0.5 and 1 mm phantoms, d' expectedly deviated from this linear relationship due to substantial detector and focal spot blurring of the phantoms. The small focal spot yielded up to 50% greater d' values than the large focal spot. For the two detectors, differences in d' did not exceed the estimated ∼7% error. Conclusions: The detectability indices predictably scaled with dose, diameter, magnification, and focal spot size and serve to validate the model. Results demonstrate statistically similar target detectability for both investigated detectors, despite differences in pixel dimensions. This CHO model provides a framework to evaluate angiographic system performance.

60 APPLIED LIFE SCIENCES↗

Sci—Fri PM: Topics — 06: The influence of regional dose sensitivity on salivary loss and recovery in the parotid gland

Purpose: The Quantitative Analyses of Normal Tissue Effects in the Clinic (QUANTEC 2010) survey of radiation dose-volume effects on salivary gland function has called for improved understanding of intragland dose sensitivity and the effectiveness of partial sparing in salivary glands. Regional dose susceptibility of sagittally- and coronally-sub-segmented parotid gland has been studied. Specifically, we examine whether individual consideration of sub-segments leads to improved prediction of xerostomia compared with whole parotid mean dose. Methods: Data from 102 patients treated for head-and-neck cancers at the BC Cancer Agency were used in this study. Whole mouth stimulated saliva was collected before (baseline), three months, and one year after cessation of radiotherapy. Organ volumes were contoured using treatment planning CT images and sub-segmented into regional portions. Both non-parametric (local regression) and parametric (mean dose exponential fitting) methods were employed. A bootstrap technique was used for reliability estimation and cross-comparison. Results: Salivary loss is described well using non-parametric and mean dose models. Parametric fits suggest a significant distinction in dose response between medial-lateral and anterior-posterior aspects of the parotid (p<0.01). Least-squares and least-median squares estimates differ significantly (p<0.00001), indicating fits may be skewed by noise or outliers. Salivary recovery exhibits a weakly arched dose response: the highest recovery is seen at intermediate doses. Conclusions: Salivary function loss is strongly dose dependent. In contrast no useful dose dependence was observed for function recovery. Regional dose dependence was observed, but may have resulted from a bias in dose distributions.

60 APPLIED LIFE SCIENCES↗

Development and evaluation of a method of calibrating medical displays based on fixed adaptation

Purpose: The purpose of this work was to develop and evaluate a new method for calibration of medical displays that includes the effect of fixed adaptation and by using equipment and luminance levels typical for a modern radiology department. Methods: Low contrast sinusoidal test patterns were derived at nine luminance levels from 2 to 600 cd/m{sup 2} and used in a two alternative forced choice observer study, where the adaptation level was fixed at the logarithmic average of 35 cd/m{sup 2}. The contrast sensitivity at each luminance level was derived by establishing a linear relationship between the ten pattern contrast levels used at every luminance level and a detectability index (d′) calculated from the fraction of correct responses. A Gaussian function was fitted to the data and normalized to the adaptation level. The corresponding equation was used in a display calibration method that included the grayscale standard display function (GSDF) but compensated for fixed adaptation. In the evaluation study, the contrast of circular objects with a fixed pixel contrast was displayed using both calibration methods and was rated on a five-grade scale. Results were calculated using a visual grading characteristics method. Error estimations in both observer studies were derived using a bootstrap method. Results: The contrast sensitivities for the darkest and brightest patterns compared to the contrast sensitivity at the adaptation luminance were 37% and 56%, respectively. The obtained Gaussian fit corresponded well with similar studies. The evaluation study showed a higher degree of equally distributed contrast throughout the luminance range with the calibration method compensated for fixed adaptation than for the GSDF. The two lowest scores for the GSDF were obtained for the darkest and brightest patterns. These scores were significantly lower than the lowest score obtained for the compensated GSDF. For the GSDF, the scores for all luminance levels were statistically separated from the average value; three were lower and two were higher. For the compensated GSDF, three of the scores could not be separated from the average value. Conclusions: An observer study using clinically relevant displays and luminance settings has demonstrated that the calibration of displays according to the GSDF causes the perceived contrast to be unevenly distributed when using displays with a high luminance range. As the luminance range increases, the perceived contrast in the dark and bright regions will be significantly lower than the perceived contrast in the middle of the luminance range. A new calibration method that includes the effect of fixed adaptation was developed and evaluated in an observer study and was found to distribute the contrast of the display more evenly throughout the grayscale than the GSDF.

60 APPLIED LIFE SCIENCES↗

Technical Note: Impact on detective quantum efficiency of edge angle determination method by International Electrotechnical Commission methodology for cardiac x-ray image detectors

Purpose: Cardiac x-ray detectors are used to acquire moving images in real-time for angiography and interventional procedures. Detective quantum efficiency (DQE) is not generally measured on these dynamic detectors; the required “for processing” image data and control of x-ray settings have not been accessible. By 2016, USA hospital physicists will have the ability to measure DQE and will likely utilize the International Electrotechnical Commission (IEC) standard for measuring DQE of dynamic x-ray imaging devices. The current IEC standard requires an image of a tilted tungsten edge test object to obtain modulation transfer function (MTF) for DQE calculation. It specifies the range of edge angles to use; however, it does not specify a preferred method to determine this angle for image analysis. The study aimed to answer the question “will my choice in method impact my results?” Four different established edge angle determination methods were compared to investigate the impact on DQE. Methods: Following the IEC standard, edge and flat field images were acquired on a cardiac flat-panel detector to calculate MTF and noise power spectrum, respectively, to determine DQE. Accuracy of the methods in determining the correct angle was ascertained using a simulated edge image with known angulations. Precision of the methods was ascertained using variability of MTF and DQE, calculated via bootstrapping. Results: Three methods provided near equal angles and the same MTF while the fourth, with an angular difference of 6%, had a MTF lower by 3% at 1.5 mm{sup −1} spatial frequency and 8% at 2.5 mm{sup −1}; corresponding DQE differences were 6% at 1.5 mm{sup −1} and 17% at 2.5 mm{sup −1}; differences were greater than standard deviations in the measurements. Conclusions: DQE measurements may vary by a significant amount, depending on the method used to determine the edge angle when following the IEC standard methodology for a cardiac x-ray detector. The most accurate and precise methods are recommended for absolute assessments and reproducible measurements, respectively.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

TU-AB-BRA-10: Prognostic Value of Intra-Radiation Treatment FDG-PET and CT Imaging Features in Locally Advanced Head and Neck Cancer

Purpose: To predict response to radiation treatment using computational FDG-PET and CT images in locally advanced head and neck cancer (HNC). Methods: 68 patients with State III-IVB HNC treated with chemoradiation were included in this retrospective study. For each patient, we analyzed primary tumor and lymph nodes on PET and CT scans acquired both prior to and during radiation treatment, which led to 8 combinations of image datasets. From each image set, we extracted high-throughput, radiomic features of the following types: statistical, morphological, textural, histogram, and wavelet, resulting in a total of 437 features. We then performed unsupervised redundancy removal and stability test on these features. To avoid over-fitting, we trained a logistic regression model with simultaneous feature selection based on least absolute shrinkage and selection operator (LASSO). To objectively evaluate the prediction ability, we performed 5-fold cross validation (CV) with 50 random repeats of stratified bootstrapping. Feature selection and model training was solely conducted on the training set and independently validated on the holdout test set. Receiver operating characteristic (ROC) curve of the pooled Result and the area under the ROC curve (AUC) was calculated as figure of merit. Results: For predicting local-regional recurrence, our model built on pre-treatment PET of lymph nodes achieved the best performance (AUC=0.762) on 5-fold CV, which compared favorably with node volume and SUVmax (AUC=0.704 and 0.449, p<0.001). Wavelet coefficients turned out to be the most predictive features. Prediction of distant recurrence showed a similar trend, in which pre-treatment PET features of lymph nodes had the highest AUC of 0.705. Conclusion: The radiomics approach identified novel imaging features that are predictive to radiation treatment response. If prospectively validated in larger cohorts, they could aid in risk-adaptive treatment of HNC.

60 APPLIED LIFE SCIENCES↗

TU-CD-304-11: Veritas 2.0: A Cloud-Based Tool to Facilitate Research and Innovation

Purpose: We introduce Veritas 2.0, a cloud-based, non-clinical research portal, to facilitate translation of radiotherapy research ideas to new delivery techniques. The ecosystem of research tools includes web apps for a research beam builder for TrueBeam Developer Mode, an image reader for compressed and uncompressed XIM files, and a trajectory log file based QA/beam delivery analyzer. Methods: The research beam builder can generate TrueBeam readable XML file either from scratch or from pre-existing DICOM-RT plans. DICOM-RT plan is first converted to XML format and then researcher can interactively modify or add control points to them. Delivered beam can be verified via reading generated images and analyzing trajectory log files. Image reader can read both uncompressed and HND-compressed XIM images. The trajectory log analyzer lets researchers plot expected vs. actual values and deviations among 30 mechanical axes. The analyzer gives an animated view of MLC patterns for the beam delivery. Veritas 2.0 is freely available and its advantages versus standalone software are i) No software installation or maintenance needed, ii) easy accessibility across all devices iii) seamless upgrades and iv) OS independence. Veritas is written using open-source tools like twitter bootstrap, jQuery, flask, and Python-based modules. Results: In the first experiment, an anonymized 7-beam DICOM-RT IMRT plan was converted to XML beam containing 1400 control points. kV and MV imaging points were inserted into this XML beam. In another experiment, a binary log file was analyzed to compare actual vs expected values and deviations among axes. Conclusions: Veritas 2.0 is a public cloud-based web app that hosts a pool of research tools for facilitating research from conceptualization to verification. It is aimed at providing a platform for facilitating research and collaboration. I am full time employee at Varian Medical systems, Palo Alto.

60 APPLIED LIFE SCIENCES↗

TU-CD-BRB-04: Automated Radiomic Features Complement the Prognostic Value of VASARI in the TCGA-GBM Dataset

Purpose: To compare the complementary prognostic value of automated Radiomic features to that of radiologist-annotated VASARI features in TCGA-GBM MRI dataset. Methods: For 96 GBM patients, pre-operative MRI images were obtained from The Cancer Imaging Archive. The abnormal tumor bulks were manually defined on post-contrast T1w images. The contrast-enhancing and necrotic regions were segmented using FAST. From these sub-volumes and the total abnormal tumor bulk, a set of Radiomic features quantifying phenotypic differences based on the tumor intensity, shape and texture, were extracted from the post-contrast T1w images. Minimum-redundancy-maximum-relevance (MRMR) was used to identify the most informative Radiomic, VASARI and combined Radiomic-VASARI features in 70% of the dataset (training-set). Multivariate Cox-proportional hazards models were evaluated in 30% of the dataset (validation-set) using the C-index for OS. A bootstrap procedure was used to assess significance while comparing the C-Indices of the different models. Results: Overall, the Radiomic features showed a moderate correlation with the radiologist-annotated VASARI features (r = −0.37 – 0.49); however that correlation was stronger for the Tumor Diameter and Proportion of Necrosis VASARI features (r = −0.71 – 0.69). After MRMR feature selection, the best-performing Radiomic, VASARI, and Radiomic-VASARI Cox-PH models showed a validation C-index of 0.56 (p = NS), 0.58 (p = NS) and 0.65 (p = 0.01), respectively. The combined Radiomic-VASARI model C-index was significantly higher than that obtained from either the Radiomic or VASARI model alone (p = <0.001). Conclusion: Quantitative volumetric and textural Radiomic features complement the qualitative and semi-quantitative annotated VASARI feature set. The prognostic value of informative qualitative VASARI features such as Eloquent Brain and Multifocality is increased with the addition of quantitative volumetric and textural features from the contrast-enhancing and necrotic tumor regions. These results should be further evaluated in larger validation cohorts.

60 APPLIED LIFE SCIENCES↗

Reducing false positives of microcalcification detection systems by removal of breast arterial calcifications

Purpose: In the past decades, computer-aided detection (CADe) systems have been developed to aid screening radiologists in the detection of malignant microcalcifications. These systems are useful to avoid perceptual oversights and can increase the radiologists’ detection rate. However, due to the high number of false positives marked by these CADe systems, they are not yet suitable as an independent reader. Breast arterial calcifications (BACs) are one of the most frequent false positives marked by CADe systems. In this study, a method is proposed for the elimination of BACs as positive findings. Removal of these false positives will increase the performance of the CADe system in finding malignant microcalcifications. Methods: A multistage method is proposed for the removal of BAC findings. The first stage consists of a microcalcification candidate selection, segmentation and grouping of the microcalcifications, and classification to remove obvious false positives. In the second stage, a case-based selection is applied where cases are selected which contain BACs. In the final stage, BACs are removed from the selected cases. The BACs removal stage consists of a GentleBoost classifier trained on microcalcification features describing their shape, topology, and texture. Additionally, novel features are introduced to discriminate BACs from other positive findings. Results: The CADe system was evaluated with and without BACs removal. Here, both systems were applied on a validation set containing 1088 cases of which 95 cases contained malignant microcalcifications. After bootstrapping, free-response receiver operating characteristics and receiver operating characteristics analyses were carried out. Performance between the two systems was compared at 0.98 and 0.95 specificity. At a specificity of 0.98, the sensitivity increased from 37% to 52% and the sensitivity increased from 62% up to 76% at a specificity of 0.95. Partial areas under the curve in the specificity range of 0.8–1.0 were significantly different between the system without BACs removal and the system with BACs removal, 0.129 ± 0.009 versus 0.144 ± 0.008 (p<0.05), respectively. Additionally, the sensitivity at one false positive per 50 cases and one false positive per 25 cases increased as well, 37% versus 51% (p<0.05) and 58% versus 67% (p<0.05) sensitivity, respectively. Additionally, the CADe system with BACs removal reduces the number of false positives per case by 29% on average. The same sensitivity at one false positive per 50 cases in the CADe system without BACs removal can be achieved at one false positive per 80 cases in the CADe system with BACs removal. Conclusions: By using dedicated algorithms to detect and remove breast arterial calcifications, the performance of CADe systems can be improved, in particular, at false positive rates representative for operating points used in screening.

60 APPLIED LIFE SCIENCES↗

WE-AB-207B-02: A Bayesian Network Approach for Joint Prediction of Tumor Control and Radiation Pneumonitis (RP) in Non-Small-Cell Lung Cancer (NSCLC)

Purpose: NSCLC radiotherapy treatment is a trade-off between controlling the tumor while limiting radiation-induced toxicities. Here we identify hierarchical biophysical relationships that could simultaneously influence both local control (LC) and RP by using an integrated Bayesian Networks (BN) approach. Methods: We studied 79 NSCLC patients treated on prospective protocol with 56 cases of LC and 21 events of RP. Beyond dosimetric information, each patient had 193 features including 12 clinical factors, 60 circulating blood cytokines before and during radiotherapy, 62 microRNAs, and 59 single-nucleotide polymorphisms (SNPs). The most relevant biophysical predictors for both LC and RP were identified using a Markov blanket local discovery algorithm and the corresponding BN was constructed using a score-learning algorithm. The area under the free-response receiver operating characteristics (AU-FROC) was used for performance evaluation. Cross-validation was employed to guard against overfitting pitfalls. Results: A BN revealing the biophysical interrelationships jointly in terms of LC and RP was developed and evaluated. The integrated BN included two SNPs, one microRNA, one clinical factor, three pre-treatment cytokines, relative changes of two cytokines between pre and during-treatment, and gEUDs of the GTV (a=-20) and lung (a=1). On cross-validation, the AUC prediction of independent LC was 0.85 (95% CI: 0.75–0.95) and RP was 0.83 (0.73–0.92). The AU-FROC of the integrated BN to predict both LC/RP was 0.81 (0.71–0.90) based on 2000 stratified bootstrap, indicating minimal loss in joint prediction power. Conclusions: We developed a new approach for multiple outcome utility application in radiotherapy based on integrated BN techniques. The BN developed from large-scale retrospective data is able to simultaneously predict LC and RP in NSCLC treatments based on individual patient characteristics. The joint prediction is only slightly compromised compared to independent predictions. Our approach shows promise for use in clinical decision support system for personalized radiotherapy subject to multiple endpoints. These studies were supported by a grant from the NCI/NIH P01-CA59827.

60 APPLIED LIFE SCIENCES↗

Impact factor for gluon production in multi-Regge kinematics in the next-to-leading order

The one-loop correction to the impact factor for gluon production upon the transition of a one-Reggeon state in the t channel to a two-Reggeon state is found. This impact factor is an element of multiparticle amplitudes in multi-Regge kinematics. The correction in question is necessary for developing the theory of Regge and multi-Regge processes. In particular, it is necessary for proving the multi-Regge form of the amplitude in the next-to-leading-logarithm approximation. This correction also makes it possible to complete the verification of the last of the unproven bootstrap conditions for gluon Reggeization and to prove, in this approximation, the validity of the multi-Regge form of the amplitude. All necessary calculations are presented, and an explicit expression for the impact factor in front of all possible color states in the t channel is given.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗