Asymptotic Source Detection Performance of Gamma-Ray Imaging Systems Under Model Mismatch

Lingenfelter, Daniel J.; Fessler, Jeffrey A.; Scott, Clayton; He, Zhong

doi:10.1109/tsp.2011.2162326

Cited by 5 publications

(36 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The simulation results of [5] do not demonstrate that the proposed method can accurately predict the performance of real detectors. In this work, we show that the performance prediction method that accounts for model mismatch developed in [5] can accurately predict source detection performance with a real Compton imaging system.…”

mentioning

confidence: 84%

“…In this work, we show that the performance prediction method that accounts for model mismatch developed in [5] can accurately predict source detection performance with a real Compton imaging system. To our knowledge, this work is the first to successfully apply an asymptotic performance prediction method to characterize the performance of the source detection task with a real gamma-ray imaging system.…”

mentioning

confidence: 91%

“…We developed asymptotic approximations for the distributions of likelihood-based estimates in [5], and used these approximations to predict detection performance in the presence of model mismatch. It was shown in [5] that the asymptotic performance prediction method yields more accurate predictions in terms of mean-square error than empirical methods, especially when few measurements are available. It was also shown in [5] that ROC predictions based on asymptotics achieve the same accuracy, in terms of mean-squared error of the area under the ROC, with fewer recorded or simulated events.…”

mentioning

confidence: 99%

“…It was shown in [5] that the asymptotic performance prediction method yields more accurate predictions in terms of mean-square error than empirical methods, especially when few measurements are available. It was also shown in [5] that ROC predictions based on asymptotics achieve the same accuracy, in terms of mean-squared error of the area under the ROC, with fewer recorded or simulated events. This reduces the computational cost of producing the ROC curve because system response calculations can be expensive.…”

mentioning

confidence: 99%

“…In addition to demonstrating the practical utility of the asymptotic performance predictions of [5], this work serves as an example application of the asymptotic performance prediction method using real data. We use the asymptotic method to predict the probability of detection as a function of scan time 0018-9499/$31.00 © 2013 IEEE with a fixed false-alarm rate for various source-to-background ratios.…”

mentioning

confidence: 99%

See 4 more Smart Citations

Source Detection Performance Prediction for a CdZnTe Array

Lingenfelter

Fessler

Scott

et al. 2013

IEEE Trans. Nucl. Sci.

Self Cite

View full text Add to dashboard Cite

Abstract-The complex system response of 3D position-sensitive gamma-ray detectors complicates modeling the recorded measurements and makes exact expressions for detection performance intractable. This makes source detection performance difficult and expensive to compute. Asymptotic analysis has the potential to simplify detection performance prediction with complex systems and has previously been applied to detection performance prediction with simulated gamma-ray detectors. In this work, we use asymptotic performance prediction methods to predict points on the receiver operating characteristic (ROC) curve for the illustrative task of detecting a 137 Cs source in background with an 18-detector CdZnTe array. We assume that the source position, source energy, background spectrum, and background spatial distribution are known. Although these assumptions are not always valid in practice, the efficiency of the prediction method makes it attractive for detection system design problems. Our results show that the asymptotic performance prediction method accurately predicts the empirically observed performance with real data recorded with a real system. Our results also characterize the performance of the detector array for the task of source detection. The accuracy and computational efficiency of the asymptotic detection performance prediction method make it a viable alternative to empirical performance evaluation.

show abstract

mentioning

confidence: 84%

mentioning

confidence: 91%

mentioning

confidence: 99%

mentioning

confidence: 99%

mentioning

confidence: 99%

See 3 more Smart Citations

Source Detection Performance Prediction for a CdZnTe Array

Lingenfelter

Fessler

Scott

et al. 2013

IEEE Trans. Nucl. Sci.

Self Cite

View full text Add to dashboard Cite

show abstract

Predicting ROC curves for source detection under model mismatch

Lingenfelter¹,

Fessler²,

Scott³

et al. 2010

IEEE Nuclear Science Symposuim &Amp; Medical Imaging Conference

View full text Add to dashboard Cite

Performance predictions and test thresholds for the task of detecting a gamma-ray source in background with a position-sensitive detector are often costly to compute empirically. The asymptotic distributions of test statistics for detecting a point-source in background give reasonable performance predictions in terms of the receiver operating characteristic curve (ROC) with less simulated or measured data than empirical methods. The asymptotic distributions also allow the user to determine the proper threshold to achieve a desired false alarm rate. Typically only approximate models are available or practical for complex gamma-ray imaging systems such as 3D positionsensitive semiconductor detectors. Applying standard formulas for the asymptotic distributions of maximum likelihood (ML) estimates in the presence of model mismatch can yield inaccurate and sometimes overly optimistic predictions of detection performance. We apply the theory of the asymptotic distribution of ML estimates under model mismatch to the case of detecting a pointsource in background with a 3D position-sensitive CdZnTe detector employing an approximate model for the system response. We show that performance predictions computed using an asymptotic approximation that accounts for mismatch more closely match the empirical performance than predictions generated by an asymptotic approximation that ignores model mismatch. I. INTRODUCTIONStatistical methods for gamma-ray source detection are often likelihood-based, meaning that the test statistic is a function of a maximum-likelihood estimate. The generalized likelihood ratio test (GLRT) is often used in the case where parameters, such as source intensity and position, are unknown [1]. The source intensity test (SIT) [2] compares the ML estimate of source intensity directly to a threshold, whereas the GLRT uses a function of the ML estimate.The performance of likelihood-based tests is often difficult to estimate because one must simulate or record a large amount of data with and without a source present to generate an empirical receiver operating characteristic curve (ROC). The ROC measures detection performance by showing the probability of detection as a function of the probability of false alarm [3]. Empirical ROC curves have been used in nuclear material detection applications, e.g., [1] and [4]. Our previous work in [2] employed asymptotic approximations to the distributions of ML estimates to compute the ROC for detecting a source in background using less data than is needed for empirical calculation, but without accounting for model mismatch.Likelihood-based tests are model-based, and their performance depends on the accuracy of the system model. Complex sensors, such as 3D position-sensitive gamma-ray detectors, are often difficult to model exactly due to the non-Gaussian uncertainties and nonlinear system response. The asymptotic

show abstract

Detection performance prediction for CdZnTe array

Lingenfelter

Fessler

Scott

et al. 2011

2011 IEEE Nuclear Science Symposium Conference Record

Self Cite

View full text Add to dashboard Cite

The complex system response of 3D position sensitive gamma-ray detectors complicates the model for the recorded measurements and makes exact expressions for detection performance intractable. This makes source detection performance difficult and expensive to com pute. Asymptotic analysis has the potential to simplify detection performance prediction with complex systems and has previously been applied to detection performance prediction with simulated gamma-ray detectors. In this work, we use asymptotic performance prediction methods to predict points on the receiver operating characteristic (ROC) curve for the illustrative task of detecting a Cs-137 source in background with an I8-detector CdZnTe array.We assume that the source position, background spectrum, and background spatial distribution are known. Our results show that the asymptotic performance prediction method accurately predicts the empirically observed performance even with real data recorded with a real system. Our results also characterize the performance of the detector array for the task of source detection. The accuracy and computational efficiency of the asymptotic detection performance prediction method make it a viable alternative to empirical performance evaluation.Gamma-ray source detection problems arise in se curity screening, nuclear nonproliferation, and medical diagnostics. Simple systems for radioactive source detec tion look for an increase in the rate of received photons due to a radiation source. More complex measurement systems use spatial and spectral information to achieve better performance, but these systems often have a com plicated system response, making it difficult to compute detection performance analytically.In this work, we quantify detection performance in terms of the receiver operating characteristic (ROC) curve, which is the probability of detection as a function of the probability of false alarm [1].that characterized the detection performance of garnma ray detectors relied on empirical ROC calculation, e.g.,[2], [3]. Empirical ROC calculation is computationally expensive and provides only limited intuition about how detector or environment parameters affect detection performance.Asymptotic ROC prediction is a computationally ef ficient alternative to empirical ROC computation for likelihood-based tests, or tests that are functions of estimates obtained by maximizing a modeled likelihood. We developed asymptotic approximations for the distri butions of likelihood-based estimates in [4], and used these approximations to predict detection performance in the presence of model mismatch. It was shown in [4] that the asymptotic performance prediction method yields more accurate predictions in terms of mean square error than empirical methods, especially when few measurements are available.The simulation results of [4] do not demonstrate that the proposed method can accurately predict the perfor mance of real detectors. In this work, we show that the performance prediction method that accounts for model mismatch develope...

show abstract

Asymptotic Source Detection Performance of Gamma-Ray Imaging Systems Under Model Mismatch

Cited by 5 publications

References 27 publications

Source Detection Performance Prediction for a CdZnTe Array

Source Detection Performance Prediction for a CdZnTe Array

Predicting ROC curves for source detection under model mismatch

Detection performance prediction for CdZnTe array

Contact Info

Product

Resources

About