A Visualization System for Performance Analysis of Image Classification Models
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.
Abstract not provided.
Disclosed herein are systems, methods, systems and devices for measurement and visualization of chemical dynamics in living cells or tissues for diagnostic pathology. Devices can be open- or closed-channel microfluidic membrane devices for long-term IR spectroscopy of live adherent cells and ultimately for rapidly identifying time-dependent spectral features indicative of chemical abnormality in individual cells.
Explore the source record for details and available documents.
Explore the source record for details and available documents.
Purpose: To identify prognostic biomarkers in pancreatic cancer using high-throughput quantitative image analysis. Methods and Materials: In this institutional review board–approved study, we retrospectively analyzed images and outcomes for 139 locally advanced pancreatic cancer patients treated with stereotactic body radiation therapy (SBRT). The overall population was split into a training cohort (n=90) and a validation cohort (n=49) according to the time of treatment. We extracted quantitative imaging characteristics from pre-SBRT {sup 18}F-fluorodeoxyglucose positron emission tomography, including statistical, morphologic, and texture features. A Cox proportional hazard regression model was built to predict overall survival (OS) in the training cohort using 162 robust image features. To avoid over-fitting, we applied the elastic net to obtain a sparse set of image features, whose linear combination constitutes a prognostic imaging signature. Univariate and multivariate Cox regression analyses were used to evaluate the association with OS, and concordance index (CI) was used to evaluate the survival prediction accuracy. Results: The prognostic imaging signature included 7 features characterizing different tumor phenotypes, including shape, intensity, and texture. On the validation cohort, univariate analysis showed that this prognostic signature was significantly associated with OS (P=.002, hazard ratio 2.74), which improved upon conventional imaging predictors including tumor volume, maximum standardized uptake value, and total legion glycolysis (P=.018-.028, hazard ratio 1.51-1.57). On multivariate analysis, the proposed signature was the only significant prognostic index (P=.037, hazard ratio 3.72) when adjusted for conventional imaging and clinical factors (P=.123-.870, hazard ratio 0.53-1.30). In terms of CI, the proposed signature scored 0.66 and was significantly better than competing prognostic indices (CI 0.48-0.64, Wilcoxon rank sum test P<1e-6). Conclusion: Quantitative analysis identified novel {sup 18}F-fluorodeoxyglucose positron emission tomography image features that showed improved prognostic value over conventional imaging metrics. If validated in large, prospective cohorts, the new prognostic signature might be used to identify patients for individualized risk-adaptive therapy.
A new image analysis procedure for quantifying microcracks from three-dimensional (3D) X-ray microCT images of concrete is presented. The method separates microcracks from air voids and aggregates by combining filtering and morphological operations. It was applied to study the effects of supplementary cementitious materials (SCMs) and curing age on autogenous shrinkage-induced microcracks in low w/b ratio concretes, and to determine the representative elementary volume (REV) for various properties of microcracks and air voids. Results showed that slag and silica fume significantly increased autogenous shrinkage and related microcracking. These SCMs increased volume fraction, width, length, dendritic density, anisotropy, and connectivity of microcracks, but decreased specific surface and tortuosity. Similar trends were observed with age. Comparison between 3D and 2D measurements was made. REV analysis showed that a sampling volume of ~20 × 20 × 25 mm{sup 3} is sufficient for characterising most parameters of autogenous shrinkage microcracks and air voids in concrete.
Synoptic images of the global cloud pattern composited from six contemporaneous satellites provide an unprecedented view of the global cloud field. Having horizontal resolution of about 0.5 deg and temporal resolution of 3 h, the global cloud imagery (GCI) resolves most of the variability of organized convection, including several harmonics of the diurnal cycle. Although the GCI has these attractive features, the dense and 3D nature of that data make it a formidable volume of information to treat in a practical and efficient manner. An interactive image-analysis system (IAS) has been developed to investigate the space-time variability of global cloud behavior. In the IAS, data, hardware, and software are integrated into a single system providing a variety of space-time covariance analyses in a menu-driven format. Owing to its customized architecture and certain homogeneous properties of the GCI, the IAS calculates such quantities effectively. Many covariance statistics are derived from 3D data with interactive speed, allowing the user to interrogate the archive iteratively in a single session. The 3D nature of those analyses and the speed with which they are performed distinguish the IAS from conventional image processing of 2D data.
We argue that many image analysis tasks can be performed by distributed amateurs. Our pilot study, with crater surveying and classification, has produced encouraging results in terms of both quantity (100,000 crater entries in 2 months) and quality. Additional information is contained in the original extended abstract.
NASA's Goddard Space Flight Center has developed a fast and effective method for generating image segmentation hierarchies. These segmentation hierarchies organize image data in a manner that makes their information content more accessible for analysis. Image segmentation enables analysis through the examination of image regions rather than individual image pixels. In addition, the segmentation hierarchy provides additional analysis clues through the tracing of the behavior of image region characteristics at several levels of segmentation detail. The potential for extracting the information content from imagery data based on segmentation hierarchies has not been fully explored for the benefit of the Earth and space science communities. This paper explores the potential of exploiting these segmentation hierarchies for the analysis of multi-date data sets, and for the particular application of change monitoring.
High Efficiency Particulate Air (HEPA) filters, commonly used in nuclear filtration applications, are an integral part of the waste management processes in nuclear plants. HEPA filters are 99.97% efficient filtration devices characterized by their high resistance to air flow, or pressure drop. From theoretical models, the initial pressure drop across a clean filter is proven to be a function of the filter media fiber diameters and porosity of the media. These filter properties are relatively difficult to obtain with traditional manual methods; therefore, there is a need to develop analytical methods to find the fiber diameter and porosity to determine the pressure drop. Typically, these values can be found by substituting a calculated representative value based upon measured values, such as equivalent fiber diameter based upon a clean pressure drop. However, these values may also be directly recorded and measured by incorporating Scanning Electron Microscope (SEM) image analysis and other measurement methods to analyze the filter media and determine its physical properties without the need for media testing. DiameterJ, an open source Java plug-in, used with ImageJ, can process images of filter media taken by an SEM to find statistical data such as mean fiber diameter and porosity. To produce the raw data, SEM images of two filter media type samples are taken and segmented in DiameterJ using the traditional and statistical region merging segmentation algorithms. Manual segmentation is necessary after the initial segmentation by the algorithms as the images tend to be too complex for the algorithms to output with the necessary accuracy. However, complications exist in the manual segmentation process as these methods can be time intensive and prone to the individual bias of the user. This in turn can skew the final mean fiber diameter result, and lead to either an over or under prediction of the pressure drop. It was also discovered that the porosity data produced by DiameterJ is inaccurate, as the SEM analyzes the three-dimensional filter media by projecting its geometry onto a plane and analyzing it as a two-dimensional binary image. Thus, the porosity is artificially inflated through the segmentation process, rendering this result incorrect. Alternatively, density determination, gravimetric analysis, and thickness testing of the filter media is collectively used to determine the filter fiber porosity. Together, the SEM image analysis and analytical lab methods produce results through direct measurements which allow for the prediction of the initial pressure drop from clean filter media without the need for prior media testing to collect pressure data. By improving upon this proposed analytical method in the future, there is potential to streamline the process of finding these filter properties into a more direct methodology for determining the pressure drop across HEPA filters.
This paper discusses the different analysis methods used in holographic particle image velocimetry to measure particle displacement and compares their relative performance. A digital holographic microscope is described and is used to record the light scattered by particles deposited on cover slides that are displaced between exposures. In this way, particle position and displacement are controlled and a numerical data set is generated. Data extraction using nearest neighbor analysis and correlation of either the reconstructed complex amplitude or intensity fields is then investigated.
The combined effects of imaging gathering, sampling and reconstruction are analyzed in terms of image fidelity. The analysis is based upon a standard end-to-end linear system model which is sufficiently general so that the results apply to most line-scan and sensor-array imaging systems. Shift-variant sampling effects are accounted for with an expected value analysis based upon the use of a fixed deterministic input scene which is randomly shifted (mathematically) relative to the sampling grid. This random sample-scene phase approach has been used successfully by the author and associates in several previous related papers.
The Synthetic Aperture Radar (SAR) sensor is widely used to record data about the ground under all atmospheric conditions. The SAR acquired images have very good resolution which necessitates the development of a classification system that process the SAR images to extract useful information for different applications. In this work, a complete system for the land cover classification was designed and programmed using the Khoros, a data flow visual language environment, taking full advantages of the polymorphic data services that it provides. Image analysis was applied to SAR images to improve and automate the processes of recognition and classification of the different regions like mountains and lakes. Both unsupervised and supervised classification utilities were used. The unsupervised classification routines included the use of several Classification/Clustering algorithms like the K-means, ISO2, Weighted Minimum Distance, and the Localized Receptive Field (LRF) training/classifier. Different texture analysis approaches such as Invariant Moments, Fractal Dimension and Second Order statistics were implemented for supervised classification of the images. The results and conclusions for SAR image classification using the various unsupervised and supervised procedures are presented based on their accuracy and performance.
Analysis of images of biotinylated Escherichia coli 70S ribosome particles, bound to streptavidin affinity grids, demonstrates that the image-quality of particles can be predicted by the image-quality of the monolayer crystalline support film. Also, the quality of the Thon rings is a good predictor of the image-quality of particles, but only when images of the streptavidin crystals extend to relatively high resolution. When the estimated resolution of streptavidin was 5 Å or worse, for example, the ribosomal density map obtained from 22,697 particles went to only 9.5 Å, while the resolution of the map reached 4.0 Å for the same number of particles, when the estimated resolution of streptavidin crystal was 4 Å or better. It thus is easy to tell which images in a data set ought to be retained for further work, based on the highest resolution seen for Bragg peaks in the computed Fourier transforms of the streptavidin component. The refined density map obtained from 57,826 particles obtained in this way extended to 3.6 Å, a marked improvement over the value of 3.9 Å obtained previously from a subset of 52,433 particles obtained from the same initial data set of 101,213 particles after 3-D classification. These results are consistent with the hypothesis that interaction with the air-water interface can damage particles when the sample becomes too thin. Finally, streptavidin monolayer crystals appear to provide a good indication of when that is the case.
Digital imaging techniques are utilized as a measure of surface displacement components in laser speckle metrology. An image scanner which is interfaced to a computer records and stores in memory the laser speckle patterns of an object in a reference and deformed configuration. Subsets of the deformed images are numerically correlated with the references as a measure of surface displacements. Discrete values are determined around a closed contour for plane problems which then become input into a boundary integral equation method in order to calculate surface traction in the contour. Stresses are then calculated within this boundary. The solution procedure is illustrated by a numerical example of a case of uniform tension.
Mathematical analysis of image sampling and interpolative reconstruction is summarized and extended to two dimensions for application to data acquired from satellite sensors such as the Thematic mapper and SPOT. It is shown that sample-scene phase influences the reconstruction of sampled images, adds a considerable blur to the average system point spread function, and decreases the average system modulation transfer function. It is also determined that the parametric bicubic interpolator with alpha = -0.5 is more radiometrically accurate than the conventional bicubic interpolator with alpha = -1, and this at no additional cost. Finally, the parametric bicubic interpolator is found to be suitable for adaptive implementation by relating the alpha parameter to the local frequency content of an image.
Computer vision algorithms for image analysis are often computationally demanding. Application of such algorithms on large image databases\---- such as the high-resolution satellite imagery covering the entire land surface, can easily saturate the computational capabilities of conventional CPUs. There is a great demand for vision algorithms running on high performance computing (HPC) architecture capable of processing petascale image data. We exploit the parallel processing capability of GPUs to present a GPU-friendly algorithm for robust and efficient detection of settlements from large-scale high-resolution satellite imagery. Feature descriptor generation is an expensive, but a key step in automated scene analysis. To address this challenge, we present GPU implementations for three different feature descriptors\-- multiscale Historgram of Oriented Gradients (HOG), Gray Level Co-Occurrence Matrix (GLCM) Contrast and local pixel intensity statistics. We perform extensive experimental evaluations of our implementation using diverse and large image datasets. Our GPU implementation of the feature descriptor algorithms results in speedups of 220 times compared to the CPU version. We present an highly efficient settlement detection system running on a multiGPU architecture capable of extracting human settlement regions from a city-scale sub-meter spatial resolution aerial imagery spanning roughly 1200 sq. kilometers in just 56 seconds with detection accuracy close to 90\%. This remarkable speedup gained by our vision algorithm maintaining high detection accuracy clearly demonstrates that such computational advancements clearly hold the solution for petascale image analysis challenges.