Search NASA⌕ Search

SEARCH · Search NASA

Results for “classifiers”

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 1,081 records · Page 60

Redefining marginal land for bioenergy crop production

Abstract Marginal land has received wide attention for its potential to produce bioenergy feedstocks while minimizing diversion of productive agricultural land from food crop production. However, there has been no consensus in the literature on how to define or identify land that is marginal for food crops and beneficial for bioenergy crops. Studies have used different definitions to quantify the amount of such land available; these have largely been based on assumed biophysical thresholds for soil quality and productivity that are unchanging over space and time. We discuss the limitations of these definitions and the rationale for considering economic returns and environmental outcomes in classifying land as marginal. We then propose the concept of “socially” marginal which is defined as land that is earning close to zero returns after accounting for the monetized costs of environmental externalities generated. We discuss a broad set of criteria for classifying land as socially marginal for food crops and suitable for bioenergy crops; with these criteria, this classification depends on spatially varying and time‐varying factors, such as climate and market conditions and policy incentives. While there are challenges related to identifying this marginal land, satellite and other large‐scale datasets increasingly enable such analysis at a fine spatial resolution. We also discuss reasons why landowners might choose not to convert bioenergy‐suitable land to bioenergy crops, and thus the need for policy incentives to support conversion of land that is socially beneficial for bioenergy crop production.

09 BIOMASS FUELS↗

The Vertebrate Breed Ontology: Toward Effective Breed Data Standardization

Abstract Background Limited universally-adopted data standards in veterinary medicine hinder data interoperability and therefore integration and comparison; this ultimately impedes the application of existing information-based tools to support advancement in diagnostics, treatments, and precision medicine. Hypothesis/Objectives A single, coherent, logic-based standard for documenting breed names in health, production, and research-related records will improve data use capabilities in veterinary and comparative medicine. Animals No live animals were used. Methods The Vertebrate Breed Ontology (VBO) was created from breed names and related information compiled from the Food and Agriculture Organization of the United Nations, breed registries, communities, and experts, using manual and computational approaches. Each breed is represented by a VBO term that includes breed information and provenance as metadata. VBO terms are classified using description logic to allow computational applications and Artificial Intelligence–readiness. Results VBO is an open, community-driven ontology representing over 19 500 livestock and companion animal breed concepts covering 49 species. Breeds are classified based on community and expert conventions (e.g., cattle breed) and supported by relations to the breed's genus and species indicated by National Center for Biotechnology Information (NCBI) Taxonomy terms. Relationships between VBO terms (e.g., relating breeds to their foundation stock) provide additional context to support advanced data analytics. VBO term metadata includes synonyms, breed identifiers/codes, and attributed cross-references to other databases. Conclusion and Clinical Importance The adoption of VBO as a standard for breed names in databases and veterinary electronic health records enhances veterinary data interoperability and computability, supporting precision medicine.

Veterinary Sciences↗

The Miller Range 090340 and 090206 meteorites: Identification of new brachinite-like achondrites with implications for the diversity and petrogenesis of the brachinite clan

Miller Range (MIL) 090340 and MIL 090206 are olivine-rich achondrites originally classified as ureilites. We investigate their petrography, mineral compositions, olivine Cr valences, equilibration temperatures, and (for MIL 090340) oxygen isotope compositions, and compare them with ureilites and other olivine-rich achondrites. We conclude that they are brachinite-like achondrites that provide new insights into the petrogenesis of brachinite clan meteorites. MIL 090340,6 has a granoblastic texture and consists of ~97 modal % by area olivine (Fo = molar Mg/[Mg+Fe] = 71.3 ± 0.6). It also contains minor to trace augite, chromite, chlorapatite, orthopyroxene, metal, troilite, and terrestrial Fe-oxides. Approximately 80% by area of MIL 090206,5 has a granoblastic texture of olivine (Fo 72.3 ± 0.1) plus minor augite and chromite, similar to MIL 090340 but also containing minor plagioclase. The rest of the section consists of a single crystal of orthopyroxene (~11 × 3 mm), poikilitically enclosing rounded grains of olivine (Fo = 76.1 ± 0.6), augite, chromite, metal, and sulfide. Equilibration temperatures for MIL 090340 and MIL 090206, calculated from olivine-spinel, olivine-augite, and two-pyroxene thermometry range from ~800 to 930 °C. In both samples, symplectic intergrowths of Ca-poor orthopyroxene + opaque phases (Fe-oxides, sulfide, metal) occur as rims on and veins/patches within olivine. Before terrestrial weathering, the opaques were probably mostly sulfide, with minor metal. All petrologic properties of MIL 090340 and MIL 090206 are consistent with those of brachinite clan meteorites, and largely distinct from those of ureilites. Oxygen isotope compositions of olivine in MIL 090340 (δ 18 O = 5.08 ± 0.30‰, δ 17 O = 2.44 ± 0.21‰, and Δ 17 O = -0.20 ± 0.12‰) are also within the range of brachinite clan meteorites, and well distinguished from ureilites. Olivine Cr valences in MIL 090340 and the granoblastic area of MIL 090206 are 2.57 ± 0.06 and 2.59 ± 0.07, respectively, similar to those of three brachinites also analyzed here (Brachina, Hughes 026, Nova 003). They are higher than those of olivine in ureilites, even those containing chromite. The valence systematics of MIL 090340, MIL 090206, and the three analyzed brachinites (lower Fo = more oxidized Cr) are consistent with previous evidence that brachinite-like parent bodies were inherently more oxidized than the ureilite parent body. The symplectic orthopyroxene + sulfide/metal assemblages in MIL 090340, MIL 090206, and many brachinite clan meteorites have superficial similarities to characteristic “reduction rims” in ureilites. However, they differ significantly in detail. They likely formed by reaction of olivine with S-rich fluids, with only minor reduction. MIL 090340 and the granoblastic area of MIL 090206 are similar in modal mineralogy and texture to most brachinites, but have higher Fo values typical of brachinite-like achondrites. The poikilitic pyroxene area of MIL 090206 is more typical of brachinite-like achondrites. The majority of their properties suggest that MIL 090340 and MIL 090206 are residues of low-degree partial melting. The poikilitic area of MIL 090206 could be a result of limited melt migration, with trapping and recrystallization of a small volume of melt in the residual matrix. Finally, these two samples are so similar in mineral compositions, Cr valence, and cosmic ray exposure ages that they could be derived from the same lithologic unit on a common parent body.

58 GEOSCIENCES↗

Compensatory reserve detects subclinical shock with more expeditious prediction for need of life‐saving interventions compared to systolic blood pressure and blood lactate

Abstract Introduction We conducted a prospective observational study on 205 trauma patients at a level I trauma facility to test the hypothesis that a compensatory reserve measurement (CRM) would identify higher risk for progression to shock and/or need a life‐saving interventions (LSIs) earlier than systolic blood pressure (SBP) and blood lactate (LAC). Methods A composite outcome metric included blood transfusion, procedural LSI, and mortality. Discrete measures assessed as abnormal (ab) were SBP <90 mmHg, CRM <60%, and LAC >2.0. A graded categorization of shock was defined as: no shock (normal [n] SBP [n‐SBP], n‐CRM, n‐LAC); sub‐clinical shock (ab‐CRM, n‐SBP, n‐LAC); occult shock (n‐SBP, ab‐CRM, ab‐LAC); or overt shock (ab‐SBP, ab‐CRM, ab‐LAC). Results Three patients displayed overt shock, 53 displayed sub‐clinical shock, and 149 displayed no shock. After incorporating lactate into the analysis, 86 patients demonstrated no shock, 25 were classified as sub‐clinical shock, 91 were classified as occult shock, and 3 were characterized as overt shock. Each shock subcategory revealed a graded increase requiring LSI and transfusion. Initial CRM was associated with progression to shock (odds ratio = 0.97; p < .001) at an earlier time than SBP or LAC. Conclusions Initial CRM uncovers a clinically relevant subset of patients who are not detected by SBP and LAC. Our results suggest CRM could be used to more expeditiously identify injured patients likely to deteriorate to shock, with requirements for blood transfusion or procedural LSI.

Convertino, Victor A.↗

Design of Digital Twin Sensing Strategies Via Predictive Modeling and Interpretable Machine Learning

This work develops a methodology for sensor placement and dynamic sensor scheduling decisions for digital twins. The digital twin data assimilation is posed as a classification problem, and predictive models are used to train optimal classification trees that represent the map from observed data to estimated digital twin states. In addition to providing a rapid digital twin updating capability, the resulting classification trees yield an interpretable mathematical representation that can be queried to inform sensor placement and sensor scheduling decisions. The proposed approach is demonstrated for a structural digital twin of a 12 ft wingspan unmanned aerial vehicle. Offline, training data are generated by simulating scenarios using predictive reduced-order models of the vehicle in a range of structural states. Furthermore, these training data can be further augmented using experimental or other historical data. In operation, the trained classifier is applied to observational data from the physical vehicle, enabling rapid adaptation of the digital twin in response to changes in structural health. Within this context, we study the performance of the optimal tree classifiers and demonstrate how they enable explainable structural assessments from sparse sensor measurements and also inform optimal sensor placement.

47 OTHER INSTRUMENTATION↗

Unsupervised Image-Based Classification of Corrosion Severity in Automobile Engine Connecting Rods

Corrosion in engine connecting rods is a critical issue in the automotive industry, potentially leading to catastrophic engine failure, monetary losses, and safety hazards. The labor shortage in the industry further emphasizes the need for fast, accurate, and automated corrosion detection methods to ensure appropriate surface treatments can be applied to restore component integrity. We present an unsupervised image-based framework for classifying corrosion severity in automobile engine connecting rods using short-wave infrared (SWIR) and telecentric grayscale imaging. We employ the structural similarity index measure (SSIM) as a dissimilarity metric and the k-medians clustering algorithm for classification. Our algorithm achieves an overall accuracy of 80.64% for SWIR images, with 100% accuracy in classifying highly corroded samples. For grayscale images, the method attains an overall accuracy of 77.42%, with 90.91% accuracy for highly corroded samples. The method’s ability to work with different imaging modalities and its high accuracy in identifying severe corrosion cases make it a promising tool for automated corrosion assessment in the automotive industry, potentially improving efficiency and safety in engine component maintenance.

42 ENGINEERING↗

Remediation of subsurface and groundwater contamination with uranium from fuel fabrication facilities at Hanau (Germany)

This paper presents aspects of site decommissioning and clearance of a former fuel fabrication facility (development and production of fuel assemblies for research reactors and HTR) at Hanau (Germany). The main pathways for environmental contamination were deposition on soil surface and topsoil and pollution of deep soil and the aquifer by waste water channel leakage. Soil excavation could be done by classical excavator techniques. An effective removal of material from the saturated zone was possible by using advanced drilling techniques. A large amount of demolished building structure and excavated soil had to be classified. Therefore the use of conveyor detector was necessary. Nearly 100000 Mg of material (excavated soil and demolished building material) were disposed of at an underground mine. A remaining volume of 700 m{sup 3} was classified as radioactive waste. Site clearance started in 2006. Groundwater remediation and monitoring is still ongoing, but has already provided excellent results by reducing the remaining Uranium considerably. (authors)

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

Computationally Efficient Training of Deep Neural Networks via Transfer Learning

Transfer learning is a highly successful approach to train deep neural networks for customized image recognition tasks. Transfer learning leverages deep learning models that are previously trained on massive datasets and re-trains the network for a novel image recognition dataset. Typically, the advantage of transfer learning has been measured in sample efficiency, but instead, we investigate the computational efficiency of transfer learning. A good pre-trained model provides features that can be used as input to a new classifier (usually the top layers of a neural network). We show that if a good pre-trained model is selected, that training a new classifier can be much more computationally efficient than training a deep neural network without transfer learning. The first step in transfer learning is to select a pre-trained model to use as a feed-forward network for generating features. Here this selection process either uses human intuition/convenience or a methodical but computationally intensive validation loop. Instead, we would prefer a method to select the pre-trained model that will produce the best transfer results for the least amount of computation. To this end, we provide a computationally efficient metric for the fit between a pre-trained model and a novel image recognition task. The better the fit, the less computation will be needed to re-train the pre-trained model for the novel task. As machine learning becomes ubiquitous, highly-accurate trained models will proliferate and computationally efficient transfer learning methods will enable rapid development of new image recognition models.

97 MATHEMATICS AND COMPUTING↗

A comparison of histopathology imaging comprehension algorithms based on multiple instance learning

Whole slide imaging (WSI), also called digital virtual microscopy, is a new imaging modality. It allows for the application of AI and machine learning methods to cancer pathology to help establish a means for the automatic diagnosis of cancer cases. However, designing machine-learning models for WSI is computationally challenging due to its required ultra-high resolution. The current state-of-the-art models use multiple instance learning (MIL). MIL is a weakly-supervised learning method in which the model uses an array of inferences from many smaller instances to make a final classification about the entire set. In the context of WSI, researchers divide the ultra-high-resolution image into many patches. The model then classifies the slide based on an array of inferences from the patches. Among several ways of making the final classification, attention-based mechanisms have resulted in superb accuracy scores. The Transformer, one attention-based algorithm, has reported substantial improvements for WSI comprehension tasks. In this project, we studied and compared several WSI comprehension algorithms. We used the following three datasets: CAMELYON16+17, TCGALung, and TCGA-Kidney. We found that attention-based MIL algorithms performed better than standard MIL algorithms for classifying WSI images, achieving a higher mean accuracy and AUC. However, none of the attention-based algorithms performed significantly better than the others, reporting accuracy scores that varied widely. Presumably, it is due to the limited availability of training samples in the data corpus. Since it is not easy to increase the samples from human subjects, some machine learning techniques like transfer learning could help mitigate this issue.

Saunders, Adam↗

Characterization of masses in digital breast tomosynthesis: Comparison of machine learning in projection views and reconstructed slices

Purpose: In digital breast tomosynthesis (DBT), quasi-three-dimensional (3D) structural information is reconstructed from a small number of 2D projection view (PV) mammograms acquired over a limited angular range. The authors developed preliminary computer-aided diagnosis (CADx) methods for classification of malignant and benign masses and compared the effectiveness of analyzing lesion characteristics in the reconstructed DBT slices and in the PVs. Methods: A data set of MLO view DBT of 99 patients containing 107 masses (56 malignant and 51 benign) was collected at the Massachusetts General Hospital with IRB approval. The DBTs were obtained with a GE prototype system which acquired 11 PVs over a 50 deg. arc. The authors reconstructed the DBTs at 1 mm slice interval using a simultaneous algebraic reconstruction technique. The region of interest (ROI) containing the mass was marked by a radiologist in the DBT volume and the corresponding ROIs on the PVs were derived based on the imaging geometry. The subsequent processes were fully automated. For classification of masses using the DBT-slice approach, the mass on each slice was segmented by an active contour model initialized with adaptive k-means clustering. A spiculation likelihood map was generated by analysis of the gradient directions around the mass margin and spiculation features were extracted from the map. The rubber band straightening transform (RBST) was applied to a band of pixels around the segmented mass boundary. The RBST image was enhanced by Sobel filtering in the horizontal and vertical directions, from which run-length statistics texture features were extracted. Morphological features including those from the normalized radial length were designed to describe the mass shape. A feature space composed of the spiculation features, texture features, and morphological features extracted from the central slice alone and seven feature spaces obtained by averaging the corresponding features from three to 19 slices centered at the central slice were compared. For classification of masses using the PV approach, a feature extraction process similar to that described above for the DBT approach was performed on the ROIs from the individual PVs. Six feature spaces obtained from the central PV alone and by averaging the corresponding features from three to 11 PVs were formed. In each feature space for either the DBT-slice or the PV approach, a linear discriminant analysis classifier with stepwise feature selection was trained and tested using a two-loop leave-one-case-out resampling procedure. Simplex optimization was used to guide feature selection automatically within the training set in each leave-one-case-out cycle. The performance of the classifiers was evaluated by the area (A{sub z}) under the receiver operating characteristic curve. Results: The test A{sub z} values from the DBT-slice approach ranged from 0.87{+-}0.03 to 0.93{+-}0.02, while those from the PV approach ranged from 0.78{+-}0.04 to 0.84{+-}0.04. The highest test A{sub z} of 0.93{+-}0.02 from the nine-DBT-slice feature space was significantly (p=0.006) better than the highest test A{sub z} of 0.84{+-}0.04 from the nine-PV feature space. Conclusion: The features of breast lesions extracted from the DBT slices consistently provided higher classification accuracy than those extracted from the PV images.

60 APPLIED LIFE SCIENCES↗

A neural network based 3D/3D image registration quality evaluator for the head-and-neck patient setup in the absence of a ground truth

Purpose: To develop a neural network based registration quality evaluator (RQE) that can identify unsuccessful 3D/3D image registrations for the head-and-neck patient setup in radiotherapy. Methods: A two-layer feed-forward neural network was used as a RQE to classify 3D/3D rigid registration solutions as successful or unsuccessful based on the features of the similarity surface near the point-of-solution. The supervised training and test data sets were generated by rigidly registering daily cone-beam CTs to the treatment planning fan-beam CTs of six patients with head-and-neck tumors. Two different similarity metrics (mutual information and mean-squared intensity difference) and two different types of image content (entire image versus bony landmarks) were used. The best solution for each registration pair was selected from 50 optimizing attempts that differed only by the initial transformation parameters. The distance from each individual solution to the best solution in the normalized parametrical space was compared to a user-defined error threshold to determine whether that solution was successful or not. The supervised training was then used to train the RQE. The performance of the RQE was evaluated using the test data set that consisted of registration results that were not used in training. Results: The RQE constructed using the mutual information had very good performance when tested using the test data sets, yielding the sensitivity, the specificity, the positive predictive value, and the negative predictive value in the ranges of 0.960-1.000, 0.993-1.000, 0.983-1.000, and 0.909-1.000, respectively. Adding a RQE into a conventional 3D/3D image registration system incurs only about 10%-20% increase of the overall processing time. Conclusions: The authors' patient study has demonstrated very good performance of the proposed RQE when used with the mutual information in identifying unsuccessful 3D/3D registrations for daily patient setup. The classifier had very good generality and required only to be trained once for each implementation. When the RQE is incorporated with an automated 3D/3D image registration system, it can improve the robustness of the system.

60 APPLIED LIFE SCIENCES↗

Concurrent segmentation of the prostate on MRI and CT via linked statistical shape models for radiotherapy planning

Purpose: Prostate gland segmentation is a critical step in prostate radiotherapy planning, where dose plans are typically formulated on CT. Pretreatment MRI is now beginning to be acquired at several medical centers. Delineation of the prostate on MRI is acknowledged as being significantly simpler to perform, compared to delineation on CT. In this work, the authors present a novel framework for building a linked statistical shape model (LSSM), a statistical shape model (SSM) that links the shape variation of a structure of interest (SOI) across multiple imaging modalities. This framework is particularly relevant in scenarios where accurate boundary delineations of the SOI on one of the modalities may not be readily available, or difficult to obtain, for training a SSM. In this work the authors apply the LSSM in the context of multimodal prostate segmentation for radiotherapy planning, where the prostate is concurrently segmented on MRI and CT. Methods: The framework comprises a number of logically connected steps. The first step utilizes multimodal registration of MRI and CT to map 2D boundary delineations of the prostate from MRI onto corresponding CT images, for a set of training studies. Hence, the scheme obviates the need for expert delineations of the gland on CT for explicitly constructing a SSM for prostate segmentation on CT. The delineations of the prostate gland on MRI and CT allows for 3D reconstruction of the prostate shape which facilitates the building of the LSSM. In order to perform concurrent prostate MRI and CT segmentation using the LSSM, the authors employ a region-based level set approach where the authors deform the evolving prostate boundary to simultaneously fit to MRI and CT images in which voxels are classified to be either part of the prostate or outside the prostate. The classification is facilitated by using a combination of MRI-CT probabilistic spatial atlases and a random forest classifier, driven by gradient and Haar features. Results: The authors acquire a total of 20 MRI-CT patient studies and use the leave-one-out strategy to train and evaluate four different LSSMs. First, a fusion-based LSSM (fLSSM) is built using expert ground truth delineations of the prostate on MRI alone, where the ground truth for the gland on CT is obtained via coregistration of the corresponding MRI and CT slices. The authors compare the fLSSM against another LSSM (xLSSM), where expert delineations of the gland on both MRI and CT are employed in the model building; xLSSM representing the idealized LSSM. The authors also compare the fLSSM against an exclusive CT-based SSM (ctSSM), built from expert delineations of the gland on CT alone. In addition, two LSSMs trained using trainee delineations (tLSSM) on CT are compared with the fLSSM. The results indicate that the xLSSM, tLSSMs, and the fLSSM perform equivalently, all of them out-performing the ctSSM. Conclusions: The fLSSM provides an accurate alternative to SSMs that require careful expert delineations of the SOI that may be difficult or laborious to obtain. Additionally, the fLSSM has the added benefit of providing concurrent segmentations of the SOI on multiple imaging modalities.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

MR/PET quantification tools: Registration, segmentation, classification, and MR-based attenuation correction

Purpose: Combined MR/PET is a relatively new, hybrid imaging modality. A human MR/PET prototype system consisting of a Siemens 3T Trio MR and brain PET insert was installed and tested at our institution. Its present design does not offer measured attenuation correction (AC) using traditional transmission imaging. This study is the development of quantification tools including MR-based AC for quantification in combined MR/PET for brain imaging. Methods: The developed quantification tools include image registration, segmentation, classification, and MR-based AC. These components were integrated into a single scheme for processing MR/PET data. The segmentation method is multiscale and based on the Radon transform of brain MR images. It was developed to segment the skull on T1-weighted MR images. A modified fuzzy C-means classification scheme was developed to classify brain tissue into gray matter, white matter, and cerebrospinal fluid. Classified tissue is assigned an attenuation coefficient so that AC factors can be generated. PET emission data are then reconstructed using a three-dimensional ordered sets expectation maximization method with the MR-based AC map. Ten subjects had separate MR and PET scans. The PET with [{sup 11}C]PIB was acquired using a high-resolution research tomography (HRRT) PET. MR-based AC was compared with transmission (TX)-based AC on the HRRT. Seventeen volumes of interest were drawn manually on each subject image to compare the PET activities between the MR-based and TX-based AC methods. Results: For skull segmentation, the overlap ratio between our segmented results and the ground truth is 85.2 ± 2.6%. Attenuation correction results from the ten subjects show that the difference between the MR and TX-based methods was <6.5%. Conclusions: MR-based AC compared favorably with conventional transmission-based AC. Quantitative tools including registration, segmentation, classification, and MR-based AC have been developed for use in combined MR/PET.

60 APPLIED LIFE SCIENCES↗

Parametrization of textural patterns in {sup 123}I-ioflupane imaging for the automatic detection of Parkinsonism

Purpose: A novel approach to a computer aided diagnosis system for the Parkinson's disease is proposed. This tool is intended as a supporting tool for physicians, based on fully automated methods that lead to the classification of{sup 123}I-ioflupane SPECT images. Methods: {sup 123}I-ioflupane images from three different databases are used to train the system. The images are intensity and spatially normalized, then subimages are extracted and a 3D gray-level co-occurrence matrix is computed over these subimages, allowing the characterization of the texture using Haralick texture features. Finally, different discrimination estimation methods are used to select a feature vector that can be used to train and test the classifier. Results: Using the leave-one-out cross-validation technique over these three databases, the system achieves results up to a 97.4% of accuracy, and 99.1% of sensitivity, with positive likelihood ratios over 27. Conclusions: The system presents a robust feature extraction method that helps physicians in the diagnosis task by providing objective, operator-independent textural information about{sup 123}I-ioflupane images, commonly used in the diagnosis of the Parkinson's disease. Textural features computation has been optimized by using a subimage selection algorithm, and the discrimination estimation methods used here makes the system feature-independent, allowing us to extend it to other databases and diseases.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

SU-E-J-235: Varian Portal Dosimetry Accuracy at Detecting Simulated Delivery Errors

Purpose: To use receiver operating characteristic (ROC) analysis to quantify the Varian Portal Dosimetry (VPD) application's ability to detect delivery errors in IMRT fields. Methods: EPID and VPD were calibrated/commissioned using vendor-recommended procedures. Five clinical plans comprising 56 modulated fields were analyzed using VPD. Treatment sites were: pelvis, prostate, brain, orbit, and base of tongue. Delivery was on a Varian Trilogy linear accelerator at 6MV using a Millenium120 multi-leaf collimator. Image pairs (VPD-predicted and measured) were exported in dicom format. Each detection test imported an image pair into Matlab, optionally inserted a simulated error (rectangular region with intensity raised or lowered) into the measured image, performed 3%/3mm gamma analysis, and saved the gamma distribution. For a given error, 56 negative tests (without error) were performed, one per 56 image pairs. Also, 560 positive tests (with error) with randomly selected image pairs and randomly selected in-field error location. Images were classified as errored (or error-free) if percent pixels with γ<κ was < (or ≥) τ. (Conventionally, κ=1 and τ=90%.) A ROC curve was generated from the 616 tests by varying τ. For a range of κ and τ, true/false positive/negative rates were calculated. This procedure was repeated for inserted errors of different sizes. VPD was considered to reliably detect an error if images were correctly classified as errored or error-free at least 95% of the time, for some κ+τ combination. Results: 20mm{sup 2} errors with intensity altered by ≥20% could be reliably detected, as could 10mm{sup 2} errors with intensity was altered by ≥50%. Errors with smaller size or intensity change could not be reliably detected. Conclusion: Varian Portal Dosimetry using 3%/3mm gamma analysis is capable of reliably detecting only those fluence errors that exceed the stated sizes. Images containing smaller errors can pass mathematical analysis, though may be detected by visual inspection. This work was not funded by Varian Oncology Systems. Some authors have other work partly funded by Varian Oncology Systems.

60 APPLIED LIFE SCIENCES↗

TU-F-18C-09: Mammogram Surveillance Using Texture Analysis for Breast Cancer Patients After Radiation Therapy

Purpose: To study the feasibility of applying cancer risk model established from treated patients to predict the risk of recurrence on follow-up mammography after radiation therapy for both ipsilateral and contralateral breast. Methods: An extensive set of textural feature functions was applied to a set of 196 Mammograms from 50 patients. 56 Mammograms from 28 patients were used as training set, 44 mammograms from 22 patients were used as test set and the rest were used for prediction. Feature functions include Histogram, Gradient, Co-Occurrence Matrix, Run-Length Matrix and Wavelet Energy. An optimum subset of the feature functions was selected by Fisher Coefficient (FO) or Mutual Information (MI) (up to top 10 features) or a method combined FO, MI and Principal Component (FMP) (up to top 30 features). One-Nearest Neighbor (1-NN), Linear Discriminant Analysis (LDA) and Nonlinear Discriminant Analysis (NDA) were utilized to build a risk model of breast cancer from the training set of mammograms at the time of diagnosis. The risk model was then used to predict the risk of recurrence from mammogram taken one year and three years after RT. Results: FPM with NDA has the best classification power in classifying the training set of the mammogram with lesions versus those without lesions. The model of FPM with NDA achieved a true positive (TP) rate of 82% compared to 45.5% of using FO with 1-NN. The best false positive (FP) rates were 0% and 3.6% in contra-lateral breast of 1-year and 3-years after RT, and 10.9% in ipsi-lateral breast of 3-years after RT. Conclusion: Texture analysis offers high dimension to differentiate breast tissue in mammogram. Using NDA to classify mammogram with lesion from mammogram without lesion, it can achieve rather high TP and low FP in the surveillance of mammogram for patient with conservative surgery combined RT.

60 APPLIED LIFE SCIENCES↗

WE-E-BRE-05: Ensemble of Graphical Models for Predicting Radiation Pneumontis Risk

Purpose: We propose a prior knowledge-based approach to construct an interaction graph of biological and dosimetric radiation pneumontis (RP) covariates for the purpose of developing a RP risk classifier. Methods: We recruited 59 NSCLC patients who received curative radiotherapy with minimum 6 month follow-up. 16 RP events was observed (CTCAE grade ≥2). Blood serum was collected from every patient before (pre-RT) and during RT (mid-RT). From each sample the concentration of the following five candidate biomarkers were taken as covariates: alpha-2-macroglobulin (α2M), angiotensin converting enzyme (ACE), transforming growth factor β (TGF-β), interleukin-6 (IL-6), and osteopontin (OPN). Dose-volumetric parameters were also included as covariates. The number of biological and dosimetric covariates was reduced by a variable selection scheme implemented by L1-regularized logistic regression (LASSO). Posterior probability distribution of interaction graphs between the selected variables was estimated from the data under the literature-based prior knowledge to weight more heavily the graphs that contain the expected associations. A graph ensemble was formed by averaging the most probable graphs weighted by their posterior, creating a Bayesian Network (BN)-based RP risk classifier. Results: The LASSO selected the following 7 RP covariates: (1) pre-RT concentration level of α2M, (2) α2M level mid- RT/pre-RT, (3) pre-RT IL6 level, (4) IL6 level mid-RT/pre-RT, (5) ACE mid-RT/pre-RT, (6) PTV volume, and (7) mean lung dose (MLD). The ensemble BN model achieved the maximum sensitivity/specificity of 81%/84% and outperformed univariate dosimetric predictors as shown by larger AUC values (0.78∼0.81) compared with MLD (0.61), V20 (0.65) and V30 (0.70). The ensembles obtained by incorporating the prior knowledge improved classification performance for the ensemble size 5∼50. Conclusion: We demonstrated a probabilistic ensemble method to detect robust associations between RP covariates and its potential to improve RP prediction accuracy. Our Bayesian approach to incorporate prior knowledge can enhance efficiency in searching of such associations from data. The authors acknowledge partial support by: 1) CREATE Medical Physics Research Training Network grant of the Natural Sciences and Engineering Research Council (Grant number: 432290) and 2) The Terry Fox Foundation Strategic Training Initiative for Excellence in Radiation Research for the 21st Century (EIRR21)

60 APPLIED LIFE SCIENCES↗