SEARCH · Engineering Papers

Results for “Feature Extraction”

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.

SourceResearch connectionFrom yearThrough year

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

At least 217 records · Page 12

All sources NASA NTRS DOE OSTI

NASA NTRS1990

Extensions of algebraic image operators: An approach to model-based vision

Researchers extend their previous research on a highly structured and compact algebraic representation of grey-level images which can be viewed as fuzzy sets. Addition and multiplication are defined for the set of all grey-level images, which can then be described as polynomials of two variables. Utilizing this new algebraic structure, researchers devised an innovative, efficient edge detection scheme. An accurate method for deriving gradient component information from this edge detector is presented. Based upon this new edge detection system researchers developed a robust method for linear feature extraction by combining the techniques of a Hough transform and a line follower. The major advantage of this feature extractor is its general, object-independent nature. Target attributes, such as line segment lengths, intersections, angles of intersection, and endpoints are derived by the feature extraction algorithm and employed during model matching. The algebraic operators are global operations which are easily reconfigured to operate on any size or shape region. This provides a natural platform from which to pursue dynamic scene analysis. A method for optimizing the linear feature extractor which capitalizes on the spatially reconfiguration nature of the edge detector/gradient component operator is discussed.

Lerner, Bao-Ting↗

DOE OSTI2024

Reconstructing Richtmyer–Meshkov instabilities from noisy radiographs using low dimensional features and attention-based neural networks

We develop an ML-based approach for density reconstruction based on transformer neural networks. This approach is demonstrated in the setting of ICF-like double shell hydrodynamic simulations wherein the parameters related to material properties and initial conditions are varied. The new method can robustly recover the complex topologies given by the Richtmyer-Meshkoff instability (RMI) from a sequence of hydrodynamic features derived from radiographic images corrupted with blur, scatter, and noise. A noise model is developed to characterize errors in extracting features from synthetic radiographs of the simulated density field. The key component of the network is a transformer encoder that acts on a sequence of features extracted from noisy radiographs. This encoder includes numerous self-attention layers that act to learn temporal dependencies in the input sequences and increase the expressiveness of the model. This approach is shown to exhibit an excellent ability to accurately recover the RMI growth rates, despite the gas-metal interface being greatly obscured by radiographic noise. Our approach can be applied in a broad array of fields involving shock physics and material science.

47 OTHER INSTRUMENTATION↗

NASA NTRSDate not supplied

Statistical Classification of Biosignature Information using Multiple Instrument Observations

The accurate identification of biosignatures (indications of life) from data taken from remote or in situ planetary exploration is one of the most important challenges in astrobiology, the interdisciplinary field examining habitability and the potential for extraterrestrial life. This study employs machine learning algorithms to optimize the identification of biosignatures, with an emphasis on those which are agnostic to a specific biochemical basis. We exploit the wealth of terrestrial data available from biogenic and abiogenic systems to enhance efficient feature prioritization. Our dataset, pulled from public databases and laboratory recorded measurements, includes elemental abundance, isotopic fractionation, and VNIR/Raman spectra The data curation process included standardization for detection limits and ranges. Subsequent feature extraction yielded detailed inputs for machine learning, including combinations of elemental content, isotopic ratios, and parameters of spectral peaks and troughs. Feature significance was evaluated across diverse machine learning methodologies, such as k-nearest neighbors, logistic regression, Random Forest, support vector machines, and Gaussian Naïve Bayes, along with a combined voting classifier. We utilized Receiver Operating Characteristic Area Under the Curve (ROC AUC) across 2,000 50% test-train splits as a robust metric of model performance. Results revealed a promising ROC AUC of 0.853 for the combined voting classifier. Removing elemental abundance data notably reduced model accuracy (13% decrease in AUC), highlighting its critical role in biosignature detection. Several other individual data features exhibited significance within their respective data types, offering additional granularity. This research fortifies the relevance of machine learning to astrobiology, potentially enhancing life detection missions by allowing algorithmic prioritization of high-interest samples for further investigation. Future work will refine data standardization, expand the dataset to include more terrestrial systems, and incorporate convolutional neural networks for spectral feature extraction. The potential for public data sharing is also under exploration, reinforcing our commitment to collective scientific advancement.

Statistical↗

DOE OSTI2017

Prognostics and Health Management of Wind Turbines -- Current Status and Future Opportunities

The global wind industry has seen tremendous growth during the past two decades. However, the industry is challenged by premature component failures, which lead to increased turbine downtime and subsequently, cost of energy for wind power. To mitigate the impacts from these failures, the wind industry has been exploring various areas for improvements ranging from product design, new materials or lubricants, to operation and maintenance (O&M) practices. Condition-based maintenance or prognostics and health management (PHM) has been explored as one enabling technology for improving O&M practices. This chapter provides a brief overview of wind turbine PHM with a focus on operational data mining and condition monitoring of drivetrains. Some future research and development opportunities in wind turbine PHM are also briefly discussed.

17 WIND ENERGY↗

NASA NTRSDate not supplied

Classifying Agnostic Biosignatures using Raman, VNIR, and Elemental Data

How can we use our current wealth of terrestrial data, encompassing biogenic and abiogenic systems, to determine the distinguishing properties of life? SCOBI (Statistical Classification of Biosignature Information) uses machine learning techniques to algorithmically identify combinations of measurements that are “indicative of life”. A set of ~1000 observations, comprising elemental abundance, isotopic fractionation, VNIR reflectance, and (in progress) Raman spectra, have been assembled from existing literature and databases. The observations cover systems classified as “indicative alive” (e.g., cells, vegetation), “indicative non-alive” (e.g., fossils, teeth), “mixed indicative” (e.g., soil, pond water), or “non-indicative” (e.g., rocks, meteorites). VNIR data was preprocessed by linear interpolation from 400-2100 nm and smoothed with a Savitzky-Golay filter. To limit the amount of Earth-biochemistry-specific (non-agnostic) information included, the first five spectral features extracted were number of peaks, number of troughs, mean reflectance, mean peak width, and broadest peak width. To help further emphasize agnostic biosignatures, Earth-specific features such as chlorophylls have been manually flagged so that feature importance with and without them can be compared. Classifiers including k-nearest neighbors (KNN), Gaussian Naïve Bayes (GNB), logistic regression (LR), random forest (RF), and support vector machine (SVM) were implemented, as was a combination voting classifier. Performance metrics included false positive rates, false negative rates, and AUC with 50-50 test/train splits (Monte Carlo simulations). Key takeaways from this stage, prior to the inclusion of Raman spectra, are (1) the overall success rate of 0.933 AUC was most heavily influenced by the elemental abundance data; and (2) VNIR reflectance had the lowest classification performance with 0.52 AUC (58% of objects correctly classified). The next steps are to complete integration of Raman spectral data and to improve the approach to pre-processing and feature extraction for both types of spectral data, such as automated baseline removal, whole spectrum matching, and dimensionality reduction.

Biosignatures↗

DOE OSTI2014

SDSS/SEGUE spectral feature analysis for stellar atmospheric parameter estimation

Large-scale and deep sky survey missions are rapidly collecting a large amount of stellar spectra, which necessitate the estimation of atmospheric parameters directly from spectra and make it feasible to statistically investigate latent principles in a large data set. We present a technique for estimating parameters T{sub eff}, log g, and [Fe/H] from stellar spectra. With this technique, we first extract features from stellar spectra using the LASSO algorithm; then, the parameters are estimated from the extracted features using the support vector regression. On a subsample of 20,000 stellar spectra from the Sloan Digital Sky Survey (SDSS) with reference parameters provided by the SDSS/SEGUE Spectroscopic Parameter Pipeline, estimation consistency are 0.007458 dex for log T{sub eff} (101.609921 K for T{sub eff}), 0.189557 dex for log g, and 0.182060 for [Fe/H], where the consistency is evaluated by mean absolute error. Prominent characteristics of the proposed scheme are sparseness, locality, and physical interpretability. In this work, each spectrum consists of 3821 fluxes, and 10, 19, and 14 typical wavelength positions are detected, respectively, for estimating T{sub eff}, log g, and [Fe/H]. It is shown that the positions are related to typical lines of stellar spectra. This characteristic is important in investigating physical indications from analysis results. Then, stellar spectra can be described by the individual fluxes on the detected positions (PD) or local integration of fluxes near them (LI). The aforementioned consistency is the result based on features described by LI. If features are described by PD, consistency is 0.009092 dex for log T{sub eff} (124.545075 K for T{sub eff}), 0.198928 dex for log g, and 0.206814 dex for [Fe/H].

79 ASTRONOMY AND ASTROPHYSICS↗

NASA NTRS1993

The 3-D image recognition based on fuzzy neural network technology

Three dimensional stereoscopic image recognition system based on fuzzy-neural network technology was developed. The system consists of three parts; preprocessing part, feature extraction part, and matching part. Two CCD color camera image are fed to the preprocessing part, where several operations including RGB-HSV transformation are done. A multi-layer perception is used for the line detection in the feature extraction part. Then fuzzy matching technique is introduced in the matching part. The system is realized on SUN spark station and special image input hardware system. An experimental result on bottle images is also presented.

Hirota, Kaoru↗

DOE OSTI2020

Electrocardiographic changes predate Parkinson’s disease onset

Autonomic nervous system involvement precedes the motor features of Parkinson’s disease (PD). Our goal was to develop a proof-of-concept model for identifying subjects at high risk of developing PD by analysis of cardiac electrical activity. We used standard 10-s electrocardiogram (ECG) recordings of 60 subjects from the Honolulu Asia Aging Study including 10 with prevalent PD, 25 with prodromal PD, and 25 controls who never developed PD. Various methods were implemented to extract features from ECGs including simple heart rate variability (HRV) metrics, commonly used signal processing methods, and a Probabilistic Symbolic Pattern Recognition (PSPR) method. Extracted features were analyzed via stepwise logistic regression to distinguish between prodromal cases and controls. Stepwise logistic regression selected four features from PSPR as predictors of PD. The final regression model built on the entire dataset provided an area under receiver operating characteristics curve (AUC) with 95% confidence interval of 0.90 [0.80, 0.99]. The five-fold cross-validation process produced an average AUC of 0.835 [0.831, 0.839]. We conclude that cardiac electrical activity provides important information about the likelihood of future PD not captured by classical HRV metrics. Machine learning applied to ECGs may help identify subjects at high risk of having prodromal PD.

59 BASIC BIOLOGICAL SCIENCES↗

DOE OSTI2014

WE-E-17A-05: Complementary Prognostic Value of CT and 18F-FDG PET Non-Small Cell Lung Cancer Tumor Heterogeneity Features Quantified Through Texture Analysis

Purpose: Previous studies have shown that CT or 18F-FDG PET intratumor heterogeneity features computed using texture analysis may have prognostic value in Non-Small Cell Lung Cancer (NSCLC), but have been mostly investigated separately. The purpose of this study was to evaluate the potential added value with respect to prognosis regarding the combination of non-enhanced CT and 18F-FDG PET heterogeneity textural features on primary NSCLC tumors. Methods: One hundred patients with non-metastatic NSCLC (stage I–III), treated with surgery and/or (chemo)radiotherapy, that underwent staging 18F-FDG PET/CT images, were retrospectively included. Morphological tumor volumes were semi-automatically delineated on non-enhanced CT using 3D SlicerTM. Metabolically active tumor volumes (MATV) were automatically delineated on PET using the Fuzzy Locally Adaptive Bayesian (FLAB) method. Intratumoral tissue density and FDG uptake heterogeneities were quantified using texture parameters calculated from co-occurrence, difference, and run-length matrices. In addition to these textural features, first order histogram-derived metrics were computed on the whole morphological CT tumor volume, as well as on sub-volumes corresponding to fine, medium or coarse textures determined through various levels of LoG-filtering. Association with survival regarding all extracted features was assessed using Cox regression for both univariate and multivariate analysis. Results: Several PET and CT heterogeneity features were prognostic factors of overall survival in the univariate analysis. CT histogram-derived kurtosis and uniformity, as well as Low Grey-level High Run Emphasis (LGHRE), and PET local entropy were independent prognostic factors. Combined with stage and MATV, they led to a powerful prognostic model (p<0.0001), with median survival of 49 vs. 12.6 months and a hazard ratio of 3.5. Conclusion: Intratumoral heterogeneity quantified through textural features extracted from both CT and FDG PET images have complementary and independent prognostic value in NSCLC.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2024

Performance Comparison of Machine Learning Models for Ultrasonic Nondestructive Evaluation of Alkali-Silica Reaction in Concrete

Alkali-silica reaction (ASR) causes concrete degradation, leading to cracking, rebar corrosion, and reduced structural integrity, which raises safety concerns. Ultrasonic nondestructive evaluation (NDE) effectively assesses concrete properties and monitors ASR progression. However, its deployment and analysis require specialized expertise and subjective interpretation. As computational power increases, artificial intelligence (AI) and machine learning (ML) algorithms are increasingly being used to automate NDE data analysis across various industries for AI-assisted automation. Regulatory agencies are adapting to this technological shift, prompting a need to evaluate current ML technologies’ capabilities and limitations in assessing concrete material properties and damage. This report presents a comparative analysis of four ML regression models for predicting concrete material damage induced by ASR expansion using long-term ultrasonic data monitoring. The models investigated include linear regression (LR), support vector regression (SVR), shallow neural networks (NN), and deep neural networks (DNN). LR, SVR, and shallow NN models use features extracted from ultrasonic signals, whereas the DNN model processes time-domain ultrasonic signals and frequency spectra directly. The study systematically compared the models’ performance from various perspectives, including model input, prediction performance, and generalization ability. The findings indicate significant variability in model performance, with some ML algorithms achieving very high or very low prediction accuracy depending on the preprocessing and feature engineering (extraction and selection) applied. Key insights include the observation that shallow ML models (LR, SVR, and shallow NNs) require meticulous preprocessing and feature extraction to achieve high accuracy. In contrast, the DNN model, although it bypasses the need for feature engineering, necessitates extensive preprocessing to mitigate noise and computational demands. The SVR model emerged as the top performer among the shallow models, and the DNN model exhibited superior performance on specific datasets but struggled with generalization across specimens from different batches. Additionally, the SVR model is sensitive to temperature variations, whereas the DNN model is robust in this regard. Using recurrent neural networks is recommended for future ASR expansion prediction studies. Recurrent neural networks’ inherent ability to capture temporal dependencies and long-term patterns makes them well suited for analyzing sequential ultrasonic monitoring data. Overall, the results and conclusions of this study could provide insights into the capabilities and effectiveness of ML when applied to ultrasonic NDE data and help identify best practices for using ML for ultrasonic NDE of concrete material properties.

36 MATERIALS SCIENCE↗

NASA NTRS2009

Highly Scalable Matching Pursuit Signal Decomposition Algorithm

Matching Pursuit Decomposition (MPD) is a powerful iterative algorithm for signal decomposition and feature extraction. MPD decomposes any signal into linear combinations of its dictionary elements or atoms . A best fit atom from an arbitrarily defined dictionary is determined through cross-correlation. The selected atom is subtracted from the signal and this procedure is repeated on the residual in the subsequent iterations until a stopping criterion is met. The reconstructed signal reveals the waveform structure of the original signal. However, a sufficiently large dictionary is required for an accurate reconstruction; this in return increases the computational burden of the algorithm, thus limiting its applicability and level of adoption. The purpose of this research is to improve the scalability and performance of the classical MPD algorithm. Correlation thresholds were defined to prune insignificant atoms from the dictionary. The Coarse-Fine Grids and Multiple Atom Extraction techniques were proposed to decrease the computational burden of the algorithm. The Coarse-Fine Grids method enabled the approximation and refinement of the parameters for the best fit atom. The ability to extract multiple atoms within a single iteration enhanced the effectiveness and efficiency of each iteration. These improvements were implemented to produce an improved Matching Pursuit Decomposition algorithm entitled MPD++. Disparate signal decomposition applications may require a particular emphasis of accuracy or computational efficiency. The prominence of the key signal features required for the proper signal classification dictates the level of accuracy necessary in the decomposition. The MPD++ algorithm may be easily adapted to accommodate the imposed requirements. Certain feature extraction applications may require rapid signal decomposition. The full potential of MPD++ may be utilized to produce incredible performance gains while extracting only slightly less energy than the standard algorithm. When the utmost accuracy must be achieved, the modified algorithm extracts atoms more conservatively but still exhibits computational gains over classical MPD. The MPD++ algorithm was demonstrated using an over-complete dictionary on real life data. Computational times were reduced by factors of 1.9 and 44 for the emphases of accuracy and performance, respectively. The modified algorithm extracted similar amounts of energy compared to classical MPD. The degree of the improvement in computational time depends on the complexity of the data, the initialization parameters, and the breadth of the dictionary. The results of the research confirm that the three modifications successfully improved the scalability and computational efficiency of the MPD algorithm. Correlation Thresholding decreased the time complexity by reducing the dictionary size. Multiple Atom Extraction also reduced the time complexity by decreasing the number of iterations required for a stopping criterion to be reached. The Course-Fine Grids technique enabled complicated atoms with numerous variable parameters to be effectively represented in the dictionary. Due to the nature of the three proposed modifications, they are capable of being stacked and have cumulative effects on the reduction of the time complexity.

Christensen, Daniel↗

NASA NTRS2006

System Would Detect Foreign-Object Damage in Turbofan Engine

A proposed data-fusion system, to be implemented mostly in software, would further process the digitized and preprocessed outputs of sensors in a turbofan engine to detect foreign-object damage (FOD) [more precisely, damage caused by impingement of such foreign objects as birds, pieces of ice, and runway debris]. The proposed system could help a flight crew to decide what, if any, response is necessary to complete a flight safely, and could aid mechanics in deciding what post-flight maintenance action might be needed. The sensory information to be utilized by the proposed system would consist of (1) the output of an accelerometer in an engine-vibration-monitoring subsystem and (2) features extracted from a gas path analysis. ["Gas path analysis" (GPA) is a term of art that denotes comprehensive analysis of engine performance derived from readings of fuel-flow meters, shaft-speed sensors, temperature sensors, and the like.] The acceleration signal would first be processed by a wavelet-transform-based algorithm, using a wavelet created for the specific purpose of finding abrupt FOD-induced changes in noisy accelerometer signals. Two additional features extracted would be the amplitude of vibration (determined via a single- frequency Fourier transform calculated at the rotational speed of the engine), and the rate of change in amplitude due to an FOD-induced rotor imbalance. This system would utilize two GPA features: the fan efficiency and the rate of change of fan efficiency with time. The selected GPA and vibrational features would be assessed by two fuzzy-logic inference engines, denoted the "Gas Path Expert" and the "Vibration Expert," respectively (see Figure 1). Each of these inference engines would generate a "possibility" distribution for occurrence of an FOD event: Each inference engine would assign, to its input information, degrees of membership, which would subsequently be transformed into basic probability assignments for the gas path and vibration components. The outputs of the inference engines would be fused by use of Dempster s combination algorithm (more precisely, an algorithm, based on the Dempster-Shafer-Yager theory of evidence, for fusing uncertain or imprecise information) to provide a reduced body of information to a human or computer decision maker. Figure 2 depicts some outputs generated in response to simulated accelerometer and GPA signals

Torso, James A.↗

NASA NTRS2016

DeepSAT: A Deep Learning Approach to Tree-Cover Delineation in 1-m NAIP Imagery for the Continental United States

High resolution tree cover classification maps are needed to increase the accuracy of current land ecosystem and climate model outputs. Limited studies are in place that demonstrates the state-of-the-art in deriving very high resolution (VHR) tree cover products. In addition, most methods heavily rely on commercial softwares that are difficult to scale given the region of study (e.g. continents to globe). Complexities in present approaches relate to (a) scalability of the algorithm, (b) large image data processing (compute and memory intensive), (c) computational cost, (d) massively parallel architecture, and (e) machine learning automation. In addition, VHR satellite datasets are of the order of terabytes and features extracted from these datasets are of the order of petabytes. In our present study, we have acquired the National Agriculture Imagery Program (NAIP) dataset for the Continental United States at a spatial resolution of 1-m. This data comes as image tiles (a total of quarter million image scenes with ~60 million pixels) and has a total size of ~65 terabytes for a single acquisition. Features extracted from the entire dataset would amount to ~8-10 petabytes. In our proposed approach, we have implemented a novel semi-automated machine learning algorithm rooted on the principles of "deep learning" to delineate the percentage of tree cover. Using the NASA Earth Exchange (NEX) initiative, we have developed an end-to-end architecture by integrating a segmentation module based on Statistical Region Merging, a classification algorithm using Deep Belief Network and a structured prediction algorithm using Conditional Random Fields to integrate the results from the segmentation and classification modules to create per-pixel class labels. The training process is scaled up using the power of GPUs and the prediction is scaled to quarter million NAIP tiles spanning the whole of Continental United States using the NEX HPC supercomputing cluster. An initial pilot over the state of California spanning a total of 11,095 NAIP tiles covering a total geographical area of 163,696 sq. miles has produced true positive rates of around 88 percent for fragmented forests and 74 percent for urban tree cover areas, with false positive rates lower than 2 percent for both landscapes.

Imagery↗

NASA NTRS2017

Planetary Crater Detection and Registration Using Marked Point Processes, Multiple Birth and Death Algorithms, and Region-Based Analysis

Because of the large variety of sensors and spacecraft collecting data, planetary science needs to integrate various multi-sensor and multi-temporal images. These multiple data represent a precious asset, as they allow the study of targets spectral responses and of changes in the surface structure; because of their variety, they also require accurate and robust registration. A new crater detection algorithm, used to extract features that will be integrated in an image registration framework, is presented. A marked point process-based method has been developed to model the spatial distribution of elliptical objects (i.e. the craters) and a birth-death Markov chain Monte Carlo method, coupled with a region-based scheme aiming at computational efficiency, is used to find the optimal configuration fitting the image. The extracted features are exploited, together with a newly defined fitness function based on a modified Hausdorff distance, by an image registration algorithm whose architecture has been designed to minimize the computational time.

Image Processing; Pattern Recognition↗

NASA NTRS2018

Planetary Crater Detection and Registration Using Marked Point Processes, Multiple Birth and Death Algorithms, and Region-Based Analysis

Image Processing:Pattern Recognition↗

DOE OSTI2016

TU-D-207B-01: A Prediction Model for Distinguishing Radiation Necrosis From Tumor Progression After Gamma Knife Radiosurgery Based On Radiomics Features From MR Images

Purpose: To develop and validate a prediction model using radiomics features extracted from MR images to distinguish radiation necrosis from tumor progression for brain metastases treated with Gamma knife radiosurgery. Methods: The images used to develop the model were T1 post-contrast MR scans from 71 patients who had had pathologic confirmation of necrosis or progression; 1 lesion was identified per patient (17 necrosis and 54 progression). Radiomics features were extracted from 2 images at 2 time points per patient, both obtained prior to resection. Each lesion was manually contoured on each image, and 282 radiomics features were calculated for each lesion. The correlation for each radiomics feature between two time points was calculated within each group to identify a subset of features with distinct values between two groups. The delta of this subset of radiomics features, characterizing changes from the earlier time to the later one, was included as a covariate to build a prediction model using support vector machines with a cubic polynomial kernel function. The model was evaluated with a 10-fold cross-validation. Results: Forty radiomics features were selected based on consistent correlation values of approximately 0 for the necrosis group and >0.2 for the progression group. In performing the 10-fold cross-validation, we narrowed this number down to 11 delta radiomics features for the model. This 11-delta-feature model showed an overall prediction accuracy of 83.1%, with a true positive rate of 58.8% in predicting necrosis and 90.7% for predicting tumor progression. The area under the curve for the prediction model was 0.79. Conclusion: These delta radiomics features extracted from MR scans showed potential for distinguishing radiation necrosis from tumor progression. This tool may be a useful, noninvasive means of determining the status of an enlarging lesion after radiosurgery, aiding decision-making regarding surgical resection versus conservative medical management.

60 APPLIED LIFE SCIENCES↗

NASA NTRS1978

Modular on-board adaptive imaging

Feature extraction involves the transformation of a raw video image to a more compact representation of the scene in which relevant information about objects of interest is retained. The task of the low-level processor is to extract object outlines and pass the data to the high-level process in a format that facilitates pattern recognition tasks. Due to the immense computational load caused by processing a 256x256 image, even a fast minicomputer requires a few seconds to complete this low-level processing. It is, therefore, necessary to consider hardware implementation of these low-level functions to achieve real-time processing speeds. The considered project had the objective to implement a system in which the continuous feature extraction process is not affected by the dynamic changes in the scene, varying lighting conditions, or object motion relative to the cameras. Due to the high bandwidth (3.5 MHz) and serial nature of the TV data, a pipeline processing scheme was adopted as the overall architecture of this system. Modularity in the system is achieved by designing circuits that are generic within the overall system.

Eskenazi, R.↗

DOE OSTI2023

An Explainable Machine-Learning Model for Compensatory Reserve Measurement: Methods for Feature Selection and the Effects of Subject Variability

Tracking vital signs accurately is critical for triaging a patient and ensuring timely therapeutic intervention. The patient’s status is often clouded by compensatory mechanisms that can mask injury severity. The compensatory reserve measurement (CRM) is a triaging tool derived from an arterial waveform that has been shown to allow for earlier detection of hemorrhagic shock. However, the deep-learning artificial neural networks developed for its estimation do not explain how specific arterial waveform elements lead to predicting CRM due to the large number of parameters needed to tune these models. Alternatively, we investigate how classical machine-learning models driven by specific features extracted from the arterial waveform can be used to estimate CRM. More than 50 features were extracted from human arterial blood pressure data sets collected during simulated hypovolemic shock resulting from exposure to progressive levels of lower body negative pressure. A bagged decision tree design using the ten most significant features was selected as optimal for CRM estimation. This resulted in an average root mean squared error in all test data of 0.171, similar to the error for a deep-learning CRM algorithm at 0.159. By separating the dataset into sub-groups based on the severity of simulated hypovolemic shock withstood, large subject variability was observed, and the key features identified for these sub-groups differed. This methodology could allow for the identification of unique features and machine-learning models to differentiate individuals with good compensatory mechanisms against hypovolemia from those that might be poor compensators, leading to improved triage of trauma patients and ultimately enhancing military and emergency medicine.

60 APPLIED LIFE SCIENCES↗