When "small" terms matter: Coupled interference features in the transport properties of cross-conjugated molecules
Explore the source record for details and available documents.
SEARCH · Search NASA
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.
Explore the source record for details and available documents.
Explore the source record for details and available documents.
Explore the source record for details and available documents.
Explore the source record for details and available documents.
Features are extractable measurements from a sample image summarizing the information content in an image and in the process providing an essential tool in image understanding. In particular, they are useful for image classification into pre-defined classes or grouping a set of image samples (also called clustering) into clusters with similar within-cluster characteristics as defined by such features. At the lowest level, features may be the intensity levels of a pixel in an image. The intensity levels of the pixels in an image may be derived from a variety of sources. For example, it can be the temperature measurement (using an infra-red camera) of the area representing the pixel or the X-ray attenuation in a given volume element of a 3-d image or it may even represent the dielectric differential in a given volume element obtained from an MIR image. At a higher level, geometric descriptors of objects of interest in a scene may also be considered as features in the image. Examples of such features are: area, perimeter, aspect ratio and other shape features, or topological features like the number of connected components, the Euler number (the number of connected components less the number of 'holes'), etc. Occupying an intermediate level in the feature hierarchy are texture features which are typically derived from a group of pixels often in a suitably defined neighborhood of a pixel. These texture features are useful not only in classification but also in the segmentation of an image into different objects/regions of interest. At the present state of our investigation, we are engaged in the task of finding a set of features associated with an object under inspection ( typically a piece of luggage or a brief case) that will enable us to detect and characterize an explosive inside, when present. Our tool of inspection is an X-Ray device with provisions for computed tomography (CT) that generate one or more (depending on the number of energy levels used) digitized 3-dimensional attenuation images with a voxel resolution of the order of one quarter of a milimeter. In the task of feature extraction and subsequent selection of an appropriate subset thereof, several important factors need to be considered. Foremost among them are: (1) Definition of the sampling unit from which the features will be extracted for the purpose of detection/ identification of the explosives. (2) The choice of features ( given the sampling unit) to be extracted that can be used to signal the existence / identity of the explosive. (3) Robustness of the computed features under different inspection conditions. To attain robustness, invariance under the transformations of translation, scaling, rotation and change of orientation is highly desirable. (4) The computational costs in the process of feature extraction, selection and their use in explosive detection/ identification In the search for extractable features, we have done a thorough literature survey with the above factors in mind and come out with a list of features that could possibly help us in meeting our objective. We are assuming that features will be based on sampling units that are single CT slices of the target. This may however change when appropriate modifications should be made to the feature extraction process. We indicate below some of the major types of features in 2- or 3-dimensional images that have been used in the literature on application of pattern recognition (PR) techniques in image understanding and are possibly pertinent to our study. In the following paragraph, we briefly indicate the motivation that guided us in the choice of these features, and identify the nature of the constraints. The principal feature types derivable from an image will be discussed in section 2. Once the features are extracted, one must select a subset of this feature set that will retain the most useful information and remove any redundant and irrelevant information that may have a detrimental effect on the classifier performance. This is discussed in section 3. Section 4 provides a brief summary.
Purpose: Increasing evidence suggests radiomics features extracted from computed tomography (CT) images may be useful in prognostic models for patients with nonsmall cell lung cancer (NSCLC). This study was designed to determine whether such features can be reproducibly obtained from cone-beam CT (CBCT) images taken using medical Linac onboard-imaging systems in order to track them through treatment. Methods: Test-retest CBCT images of ten patients previously enrolled in a clinical trial were retrospectively obtained and used to determine the concordance correlation coefficient (CCC) for 68 different texture features. The volume dependence of each feature was also measured using the Spearman rank correlation coefficient. Features with a high reproducibility (CCC > 0.9) that were not due to volume dependence in the patient test-retest set were further examined for their sensitivity to differences in imaging protocol, level of scatter, and amount of motion by using two phantoms. The first phantom was a texture phantom composed of rectangular cartridges to represent different textures. Features were measured from two cartridges, shredded rubber and dense cork, in this study. The texture phantom was scanned with 19 different CBCT imagers to establish the features’ interscanner variability. The effect of scatter on these features was studied by surrounding the same texture phantom with scattering material (rice and solid water). The effect of respiratory motion on these features was studied using a dynamic-motion thoracic phantom and a specially designed tumor texture insert of the shredded rubber material. The differences between scans acquired with different Linacs and protocols, varying amounts of scatter, and with different levels of motion were compared to the mean intrapatient difference from the test-retest image set. Results: Of the original 68 features, 37 had a CCC >0.9 that was not due to volume dependence. When the Linac manufacturer and imaging protocol were kept consistent, 4–13 of these 37 features passed our criteria for reproducibility more than 50% of the time, depending on the manufacturer-protocol combination. Almost all of the features changed substantially when scatter material was added around the phantom. For the dense cork, 23 features passed in the thoracic scans and 11 features passed in the head scans when the differences between one and two layers of scatter were compared. Using the same test for the shredded rubber, five features passed the thoracic scans and eight features passed the head scans. Motion substantially impacted the reproducibility of the features. With 4 mm of motion, 12 features from the entire volume and 14 features from the center slice measurements were reproducible. With 6–8 mm of motion, three features (Laplacian of Gaussian filtered kurtosis, gray-level nonuniformity, and entropy), from the entire volume and seven features (coarseness, high gray-level run emphasis, gray-level nonuniformity, sum-average, information measure correlation, scaled mean, and entropy) from the center-slice measurements were considered reproducible. Conclusions: Some radiomics features are robust to the noise and poor image quality of CBCT images when the imaging protocol is consistent, relative changes in the features are used, and patients are limited to those with less than 1 cm of motion.
Purpose: Sensitivity of PET-derived texture features to reconstruction methods has been reported for features extracted from axial planes; however, studies often utilize three dimensional techniques. This work aims to quantify the impact of multi-plane (3D) vs. single-plane (2D) feature extraction on radiomics-based analysis, including sensitivity to reconstruction parameters and potential loss of spatial information. Methods: Twenty-three patients with solid tumors underwent [{sup 18}F]FDG PET/CT scans under identical protocols. PET data were reconstructed using five sets of reconstruction parameters. Tumors were segmented using an automatic, in-house algorithm robust to reconstruction variations. 50 texture features were extracted using two Methods: 2D patches along axial planes and 3D patches. For each method, sensitivity of features to reconstruction parameters was calculated as percent difference relative to the average value across reconstructions. Correlations between feature values were compared when using 2D and 3D extraction. Results: 21/50 features showed significantly different sensitivity to reconstruction parameters when extracted in 2D vs 3D (wilcoxon α<0.05), assessed by overall range of variation, Rangevar(%). Eleven showed greater sensitivity to reconstruction in 2D extraction, primarily first-order and co-occurrence features (average Rangevar increase 83%). The remaining ten showed higher variation in 3D extraction (average Range{sub var}increase 27%), mainly co-occurence and greylevel run-length features. Correlation of feature value extracted in 2D and feature value extracted in 3D was poor (R<0.5) in 12/50 features, including eight co-occurrence features. Feature-to-feature correlations in 2D were marginally higher than 3D, ∣R∣>0.8 in 16% and 13% of all feature combinations, respectively. Larger sensitivity to reconstruction parameters were seen for inter-feature correlation in 2D(σ=6%) than 3D (σ<1%) extraction. Conclusion: Sensitivity and correlation of various texture features were shown to significantly differ between 2D and 3D extraction. Additionally, inter-feature correlations were more sensitive to reconstruction variation using single-plane extraction. This work highlights a need for standardized feature extraction/selection techniques in radiomics.
Purpose: To evaluate the reproducibility of radiomics features by repeating computed tomographic (CT) scans in rectal cancer. To choose stable radiomics features for rectal cancer. Methods: 40 rectal cancer patients were enrolled in this study, each of whom underwent two CT scans within average 8.7 days (5 days to 17 days), before any treatment was delivered. The rectal gross tumor volume (GTV) was distinguished and segmented by an experienced oncologist in both CTs. Totally, more than 2000 radiomics features were defined in this study, which were divided into four groups (I: GLCM, II: GLRLM III: Wavelet GLCM and IV: Wavelet GLRLM). For each group, five types of features were extracted (Max slice: features from the largest slice of target images, Max value: features from all slices of target images and choose the maximum value, Min value: minimum value of features for all slices, Average value: average value of features for all slices, Matrix sum: all slices of target images translate into GLCM and GLRLM matrices and superpose all matrices, then extract features from the superposed matrix). Meanwhile a LOG (Laplace of Gauss) filter with different parameters was applied to these images. Concordance correlation coefficients (CCC) and inter-class correlation coefficients (ICC) were calculated to assess the reproducibility. Results: 403 radiomics features were extracted from each type of patients’ medical images. Features of average type are the most reproducible. Different filters have little effect for radiomics features. For the average type features, 253 out of 403 features (62.8%) showed high reproducibility (ICC≥0.8), 133 out of 403 features (33.0%) showed medium reproducibility (0.8≥ICC≥0.5) and 17 out of 403 features (4.2%) showed low reproducibility (ICC≥0.5). Conclusion: The average type radiomics features are the most stable features in rectal cancer. Further analysis of these features of rectal cancer can be warranted for treatment monitoring and prognosis prediction.
Purpose: Normal lung CT texture features have been used for the prediction of radiation-induced lung disease (radiation pneumonitis and radiation fibrosis). For these features to be clinically useful, they need to be relatively invariant (robust) to tumor size and not correlated with normal lung volume. Methods: The free-breathing CTs of 14 lung SBRT patients were studied. Different sizes of GTVs were simulated with spheres placed at the upper lobe and lower lobe respectively in the normal lung (contralateral to tumor). 27 texture features (9 from intensity histogram, 8 from grey-level co-occurrence matrix [GLCM] and 10 from grey-level run-length matrix [GLRM]) were extracted from [normal lung-GTV]. To measure the variability of a feature F, the relative difference D=|Fref -Fsim|/Fref*100% was calculated, where Fref was for the entire normal lung and Fsim was for [normal lung-GTV]. A feature was considered as robust if the largest non-outlier (Q3+1.5*IQR) D was less than 5%, and considered as not correlated with normal lung volume when their Pearson correlation was lower than 0.50. Results: Only 11 features were robust. All first-order intensity-histogram features (mean, max, etc.) were robust, while most higher-order features (skewness, kurtosis, etc.) were unrobust. Only two of the GLCM and four of the GLRM features were robust. Larger GTV resulted greater feature variation, this was particularly true for unrobust features. All robust features were not correlated with normal lung volume while three unrobust features showed high correlation. Excessive variations were observed in two low grey-level run features and were later identified to be from one patient with local lung diseases (atelectasis) in the normal lung. There was no dependence on GTV location. Conclusion: We identified 11 robust normal lung CT texture features that can be further examined for the prediction of radiation-induced lung disease. Interestingly, low grey-level run features identified normal lung diseases. This work was supported in part by the National Cancer Institute Grants R01CA172638.
Purpose: To determine how radiomics features change during radiation therapy and whether those changes (delta-radiomics features) can improve prognostic models built with clinical factors. Methods: 62 radiomics features, including histogram, co-occurrence, run-length, gray-tone difference, and shape features, were calculated from pretreatment and weekly intra-treatment CTs for 107 stage III NSCLC patients (5–9 images per patient). Image preprocessing for each feature was determined using the set of pretreatment images: bit-depth resample and/or a smoothing filter were tested for their impact on volume-correlation and significance of each feature in univariate cox regression models to maximize their information content. Next, the optimized features were calculated from the intratreatment images and tested in linear mixed-effects models to determine which features changed significantly with dose-fraction. The slopes in these significant features were defined as delta-radiomics features. To test their prognostic potential multivariate cox regression models were fitted, first using only clinical features and then clinical+delta-radiomics features for overall-survival, local-recurrence, and distant-metastases. Leave-one-out cross validation was used for model-fitting and patient predictions. Concordance indices(c-index) and p-values for the log-rank test with patients stratified at the median were calculated. Results: Approximately one-half of the 62 optimized features required no preprocessing, one-fourth required smoothing, and one-fourth required smoothing and resampling. From these, 54 changed significantly during treatment. For overall-survival, the c-index improved from 0.52 for clinical factors alone to 0.62 for clinical+delta-radiomics features. For distant-metastases, the c-index improved from 0.53 to 0.58, while for local-recurrence it did not improve. Patient stratification significantly improved (p-value<0.05) for overallsurvival and distant-metastases when delta-radiomics features were included. The delta-radiomics versions of autocorrelation, kurtosis, and compactness were selected most frequently in leave-one-out iterations. Conclusion: Weekly changes in radiomics features can potentially be used to evaluate treatment response and predict patient outcomes. High-risk patients could be recommended for dose escalation or consolidation chemotherapy. This project was funded in part by grants from the National Cancer Institute (NCI) and the Cancer Prevention Research Institute of Texas (CPRIT).
This report describes the author’s research toward creating algorithms for selecting useful statistical features in pattern/target classification problems in which the features are non-Gaussian distributed. In engineering practice, it is common to either (1) Not perform any feature selection procedure, or (2) Use a feature selection algorithm that assumes the features are Gaussian distributed. This process can be far from optimal if the data are actually non-Gaussian, as they frequently are in practice. This research has the goal of mitigating that problem by creating a useful algorithm that can be used in practice. The approach is to focus on the performance indices used in common feature selection algorithms, such as Branch and Bound algorithms and the Sequential Forward Selection Algorithm. Ordinarily, the performance index used to measure the class separation in feature space involves assuming the data are Gaussian and deriving tractable performance indices that use closed form equations that represent the probability density functions of the class data. The advantage of this approach is that it produces feature selection algorithms that have low computational complexity and do not require knowledge of the actual data densities. The disadvantage is that these algorithms may not perform reasonably when the data are non-Gaussian, as they commonly are in practice. This research examines the use of information-theoretic class separability measures that can deal with the non-Gaussian case. In particular, this work shows that the Hellinger Distance (a divergence measure), has very desirable mathematical properties and can be useful for feature selection. The work in this paper employs a multivariate kernel density estimator along with the Hellinger Distance and an optimal subset selection algorithm to do feature selection. This report (1) Defines the classification and feature selection problems, (2) Describes the continuous variable Hellinger distance in one dimension and discusses its mathematical properties in comparison with other distance measures, (3) Develops the discrete variable Hellinger distance expression for one and two dimensions, and shows that the discrete Hellinger distance for multiple dimensions is a straightforward extension, and (4) Presents experimental density estimation results using a new MATLAB implementation of a multivariate kernel density estimator. The draft of this report served as a starting point for Ensign Matthew Wilder’s MSEE thesis work at the Naval Postgraduate School (NPS) in 2010 and 2011. Ensign Wilder completed his thesis and graduated from NPS in June 2011 (M. J. Wilder, “Automatic Target Recognition: Statistical Feature Selection of Non-Gaussian Distributed Target Classes,” MSEE Thesis, Naval Postgraduate School, Thesis Advisor: Grace A. Clark, Second Reader: Monique P. Fargues, June 2011). In the thesis, Ensign Wilder created implementations of the Branch and Bound and Sequential Feature Selection algorithms, extended the multivariate kernel density estimator code, and evaluated the performance of Hellinger Distance-based feature selection algorithms using both simulated data and data from a real-world benchmark data set. The new algorithm was shown to be very effective for both Gaussian and non-Gaussian data sets. The advantages gained for non-Gaussian data come at the cost of increased computational complexity. Using a laptop computer for practical problems, we generally use initial feature vectors containing no more than about ten or twelve features. Proposed future research lies in creating faster kernel density estimator algorithms that would allow for the use of higher dimensional data.
A primary goal of geomorphological enquiry is to make genetic associations between process and form. In rock breakdown studies, the links between process, inheritance and lithology are not well constrained. In particular, there is a need to establish an understanding of feature persistence. That is, to determine the extent to which in situ rock breakdown (e.g., aeolian abrasion or salt weathering) masks signatures of earlier geomorphic transport processes (e.g., fluvial transport or crater ejecta). Equally important is the extent to which breakdown during geomorphic transport masks the imprint of past weathering. The use of rock features in this way raises the important question: Can features on the surface of a rock reliably indicate its geomorphic history? This has not been determined for rock surfaces on Earth or other planets. A first step towards constraining the links between process, inheritance, and morphology is to identify pristine features produced by different process regimes. The purpose of this atlas is to provide a comprehensive image collection of breakdown features commonly observed on boulders in different geomorphic environments. The atlas is intended as a tool for planetary geoscientists and their students to assist in identifying features found on rocks on planetary surfaces. In compiling this atlas, we have attempted to include features that have formed 'recently' and where the potential for modification by another geomorphic process is low. However, we acknowledge that this is, in fact, difficult to achieve when selecting rocks in their natural environment. We group breakdown features according to their formative environment and process. In selecting images for inclusion in the atlas we were mindful to cover a wide range of climatic zones. For example, in the weathering chapter, clast features are shown from locations such as the hyper-arid polar desert of Antarctica and the semi-arid canyons of central Australia. This is important as some features (e.g., alveoli) occur across climate regimes. We have drawn on the published geomorphological literature and our own field experience. We use, where possible, images of extrusive igneous rocks as the data returned from Mars, Venus and the Moon indicates that this is the predominant rock type. One of the purposes of this atlas is to expand the range of surface features that are known to indicate a particular geomorphic environment or process history. The surface features on boulders in some environments such as aeolian and weathering are well understood. In contrast, those in fluvial or ejecta environments are not. Therefore we have presented a comprehensive assemblage of features that are likely to be produced in each of the geomorphic environments. We hope that this atlas will trigger more research on diagnostic features, particularly their morphometry and detailed morphology, their persistence and rates of formation. In this first edition of the atlas we detail the features found on clasts in three geomorphic environments: aeolian, fluvial and weathering. Future editions of the atlas will include chapters on ejecta, micro-impacts, coastal, colluvial, glacial and structural features.
Purpose: Quantitative imaging is a fast evolving discipline where a large number of features are extracted from images; i.e., radiomics. Some features have been shown to have diagnostic, prognostic and predictive value. However, they are sensitive to acquisition and processing factors; e.g., noise. In this study noise was added to positron emission tomography (PET) images to determine how features were affected by noise. Methods: Three levels of Gaussian noise were added to 8 lung cancer patients PET images acquired in 3D mode (static) and using respiratory tracking (4D); for the latter images from one of 10 phases were used. A total of 62 features: 14 shape, 19 intensity (1stO), 18 GLCM textures (2ndO; from grey level co-occurrence matrices) and 11 RLM textures (2ndO; from run-length matrices) features were extracted from segmented tumors. Dimensions of GLCM were 256×256, calculated using 3D images with a step size of 1 voxel in 13 directions. Grey levels were binned into 256 levels for RLM and features were calculated in all 13 directions. Results: Feature variation generally increased with noise. Shape features were the most stable while RLM were the most unstable. Intensity and GLCM features performed well; the latter being more robust. The most stable 1stO features were compactness, maximum and minimum length, standard deviation, root-mean-squared, I30, V10-V90, and entropy. The most stable 2ndO features were entropy, sum-average, sum-entropy, difference-average, difference-variance, difference-entropy, information-correlation-2, short-run-emphasis, long-run-emphasis, and run-percentage. In general, features computed from images from one of the phases of 4D scans were more stable than from 3D scans. Conclusion: This study shows the need to characterize image features carefully before they are used in research and medical applications. It also shows that the performance of features, and thereby feature selection, may be assessed in part by noise analysis.
A method of automated recognition of three-dimensional (3D) features in images generated by ground-penetrating imaging radar (GPIR) is undergoing development. GPIR 3D images can be analyzed to detect and identify such subsurface features as pipes and other utility conduits. Until now, much of the analysis of GPIR images has been performed manually by expert operators who must visually identify and track each feature. The present method is intended to satisfy a need for more efficient and accurate analysis by means of algorithms that can automatically identify and track subsurface features, with minimal supervision by human operators. In this method, data from multiple sources (for example, data on different features extracted by different algorithms) are fused together for identifying subsurface objects. The algorithms of this method can be classified in several different ways. In one classification, the algorithms fall into three classes: (1) image-processing algorithms, (2) feature- extraction algorithms, and (3) a multiaxis data-fusion/pattern-recognition algorithm that includes a combination of machine-learning, pattern-recognition, and object-linking algorithms. The image-processing class includes preprocessing algorithms for reducing noise and enhancing target features for pattern recognition. The feature-extraction algorithms operate on preprocessed data to extract such specific features in images as two-dimensional (2D) slices of a pipe. Then the multiaxis data-fusion/ pattern-recognition algorithm identifies, classifies, and reconstructs 3D objects from the extracted features. In this process, multiple 2D features extracted by use of different algorithms and representing views along different directions are used to identify and reconstruct 3D objects. In object linking, which is an essential part of this process, features identified in successive 2D slices and located within a threshold radius of identical features in adjacent slices are linked in a directed-graph data structure. Relative to past approaches, this multiaxis approach offers the advantages of more reliable detections, better discrimination of objects, and provision of redundant information, which can be helpful in filling gaps in feature recognition by one of the component algorithms. The image-processing class also includes postprocessing algorithms that enhance identified features to prepare them for further scrutiny by human analysts (see figure). Enhancement of images as a postprocessing step is a significant departure from traditional practice, in which enhancement of images is a preprocessing step.
Spitzer Space Telescope mid-infrared spectroscopy has been obtained for 15 carbon-rich protoplanetary nebulae (PPNe) in the Large Magellanic Cloud (LMC) and for two other such stars in the Small Magellanic Cloud (SMC). Of these 17 PPNe, the unidentified 21 {mu}m feature is strong in 7 spectra, weak in 2 spectra, and very weak or questionable in 4 spectra. Two of the four spectra without the 21 {mu}m feature have a very strong feature near 11 {mu}m, similar to a feature observed in some carbon-rich planetary nebulae (PNe) in the LMC. We attribute this feature to unusual SiC dust, although the feature-to-continuum ratio is much larger than for SiC features in Galactic or Magellanic Cloud carbon star spectra. The remaining two objects show typical carbon-rich PPNe spectra with no 21 {mu}m features. One of the LMC objects that lacks the 21 {mu}m feature and one SMC object with a questionable 21 {mu}m detection may have mixed dust chemistries based upon their spectral similarity to Galactic [WC] PNe. The 13 objects that either definitely or may show the 21 {mu}m feature have distinct dust shell properties compared to the Galactic 21 {mu}m objects-the 21 {mu}m features are weaker, the estimated dust temperatures are significantly higher, the unidentified infrared (UIR) bands are stronger, and the UIRs show more structure. Four of the 21 {mu}m objects appear to show normal SiC emission features in their spectra. Many of the PPNe show strong 30 {mu}m features, although this feature carries less of the total mid-infrared emission than is normally the case for the Galactic 21 {mu}m PPNe. The LMC objects are in the LMC halo rather than in the LMC bar. The estimated luminosities of these PPNe vary from 4700 to 12,500 L{sub sun}.
Purpose: Temporal comparison of lesions might improve classification between benign and malignant lesions in full-field digital mammograms (FFDM). The authors compare the use of volumetric features for lesion classification, which are computed from dense tissue thickness maps, to the use of mammographic lesion area. Use of dense tissue thickness maps for lesion characterization is advantageous, since it results in lesion features that are invariant to acquisition parameters. Methods: The dataset used in the analysis consisted of 60 temporal mammogram pairs comprising 120 mediolateral oblique or craniocaudal views with a total of 65 lesions, of which 41 were benign and 24 malignant. The authors analyzed the performance of four volumetric features, area, and four other commonly used features obtained from temporal mammogram pairs, current mammograms, and prior mammograms. The authors evaluated the individual performance of all features and of different feature sets. The authors used linear discriminant analysis with leave-one-out cross validation to classify different feature sets. Results: Volumetric features from temporal mammogram pairs achieved the best individual performance, as measured by the area under the receiver operating characteristic curve (A{sub z} value). Volume change (A{sub z} = 0.88) achieved higher A{sub z} value than projected lesion area change (A{sub z} = 0.78) in the temporal comparison of lesions. Best performance was achieved with a set that consisted of a set of features extracted from the current exam combined with four volumetric features representing changes with respect to the prior mammogram (A{sub z} = 0.90). This was significantly better (p = 0.005) than the performance obtained using features from the current exam only (A{sub z} = 0.77). Conclusions: Volumetric features from temporal mammogram pairs combined with features from the single exam significantly improve discrimination of benign and malignant lesions in FFDM mammograms compared to using only single exam features. In the comparison with prior mammograms, use of volumetric change may lead to better performance than use of lesion area change.
Purpose: Radiation induced lung damage (RILD) is an important dose-limiting toxicity for patients treated with radiation therapy. Scoring systems for RILD are subjective and limit our ability to find robust predictors of toxicity. We investigate the dose and time-related response for texture-based lung CT image features that serve as potential quantitative measures of RILD. Methods: Pre- and post-RT diagnostic imaging studies were collected for retrospective analysis of 21 patients treated with photon or proton radiotherapy for NSCLC. Total lung and selected isodose contours (0–5, 5–15, 15–25Gy, etc.) were deformably registered from the treatment planning scan to the pre-RT and available follow-up CT studies for each patient. A CT image analysis framework was utilized to extract 3698 unique texture-based features (including co-occurrence and run length matrices) for each region of interest defined by the isodose contours and the total lung volume. Linear mixed models were fit to determine the relationship between feature change (relative to pre-RT), planned dose and time post-RT. Results: Seventy-three follow-up CT scans from 21 patients (median: 3 scans/patient) were analyzed to describe CT image feature change. At the p=0.05 level, dose affected feature change in 2706 (73.1%) of the available features. Similarly, time affected feature change in 408 (11.0%) of the available features. Both dose and time were significant predictors of feature change in a total of 231 (6.2%) of the extracted image features. Conclusion: Characterizing the dose and time-related response of a large number of texture-based CT image features is the first step toward identifying objective measures of lung toxicity necessary for assessment and prediction of RILD. There is evidence that numerous features are sensitive to both the radiation dose and time after RT. Beyond characterizing feature response, further investigation is warranted to determine the utility of these features as surrogates of clinically significant lung injury.
In data-driven automated fault detection and diagnostics (AFDD) modeling for building energy systems, feature engineering is a critical process of extracting information from high-dimensional and noisy sensor measurement and turning it into informative and representative inputs or features for data-driven modeling. However, few studies specifically discuss the feature engineering, especially the interactions between feature extraction and feature selection in whole-building AFDD. We developed a systematic feature extraction and selection framework for whole-building AFDD. In this framework, features are aggressively extracted from raw sensor data using statistical feature extraction techniques with various window sizes and statistics. With many features extracted, a hybrid feature selection algorithm that combines the filter and wrapper method then selects the best feature set. The framework considers diversity in the duration of fault behavior among fault types in whole-building AFDD, thus achieving high model generalization. We implemented our developed framework in a virtual testbed calibrated with measured data from Oak Ridge National Laboratory's Flexible Research Platform designed to mimic the operation of a typical small commercial building. The AFDD model is trained by the simulation data generated from the virtual testbed. The results show that (1) the developed framework improves the generalization of the AFDD model by 10.7% compared with literature-reported feature extraction and selection methods and (2) features with diverse window sizes and statistics are selected, providing insight into physical systems beyond the current understanding of buildings and faults and improving the detection and diagnostics of multiple fault types.