Search NASA⌕ Search

SEARCH · Search NASA

Results for “Low SNR object detection”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

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

At least 19 records

Remote sensing detection enhancement

Big Data in the area of Remote Sensing has been growing rapidly. Remote sensors are used in surveillance, security, traffic, environmental monitoring, and autonomous sensing. Real-time detection of small moving targets using a remote sensor is an ongoing, challenging problem. Since the object is located far away from the sensor, the object often appears too small. The object’s signal-to-noise-ratio (SNR) is often very low. Occurrences such as camera motion, moving backgrounds (e.g., rustling leaves), low contrast and resolution of foreground objects makes it difficult to segment out the targeted moving objects of interest. Due to the limited appearance of the target, it is tough to obtain the target’s characteristics such as its shape and texture. Without these characteristics, filtering out false detections can be a difficult task. Detecting these targets, would often require the detector to operate under a low detection threshold. However, lowering the detection threshold could lead to an increase of false alarms. In this paper, the author will introduce a new method that improves the probability to detect low SNR objects, while decreasing the number of false alarms as compared to using the traditional baseline detection technique.

47 OTHER INSTRUMENTATION↗

Remote Sensing Low Signal-to-Noise-Ratio Target Detection Enhancement

In real-time remote sensing application, frames of data are continuously flowing into the processing system. The capability of detecting objects of interest and tracking them as they move is crucial to many critical surveillance and monitoring missions. Detecting small objects using remote sensors is an ongoing, challenging problem. Since object(s) are located far away from the sensor, the target’s Signal-to-Noise-Ratio (SNR) is low. The Limit of Detection (LOD) for remote sensors is bounded by what is observable on each image frame. In this paper, we present a new method, a “Multi-frame Moving Object Detection System (MMODS)”, to detect small, low SNR objects that are beyond what a human can observe in a single video frame. This is demonstrated by using simulated data where our technology-detected objects are as small as one pixel with a targeted SNR, close to 1:1. We also demonstrate a similar improvement using live data collected with a remote camera. The MMODS technology fills a major technology gap in remote sensing surveillance applications for small target detection. Our method does not require prior knowledge about the environment, pre-labeled targets, or training data to effectively detect and track slow- and fast-moving targets, regardless of the size or the distance.

47 OTHER INSTRUMENTATION↗

Filtering Image Records Using Wavelets and the Zakai Equation

Consider the problem of detecting and localizing a faint object moving In an "essentially stationary" background, using a sequence of two-dimensional low-SNR images of the scene. A natural approach consists of "digitizing" each snapshot into a discrete set of observations, sufficiently (perhaps not exactly) matched to the object In question, then tracking the object using an appropriate stochastic filter. The tracking would be expected to make up for the low signal-to-noise ratio, this allowing one to "coherently" process successive images in order to beat down the noise and localize the object. Thus, "tracking" here does not refer to the usual notion of detecting then tracking: rather, we track in order to detect The problem then becomes one of choosing the appropriate image representation as well as the optimal (and necessarily nonlinear filter. We propose exact and approximate solutions using wavelets and the Zakai equation. The smoothness of the wavelets used is required in the derivation of the evolution equation for the conditional density giving the filter, and their orthogonality makes it possible to carry out actual computations of the Ito- and change-of-gauge-terms in the algorithm effectively.

Haddad, Ziad S.↗

ON THE NATURE OF THE TERTIARY COMPANION TO FW TAU: ALMA CO OBSERVATIONS AND SED MODELING

It is thought that planetary mass companions may form through gravitational disk instabilities or core accretion. Identifying such objects in the process of formation would provide the most direct test for the competing formation theories. One of the most promising candidates for a planetary mass object still in formation is the third object in the FW Tau system. We present here ALMA cycle 1 observations confirming the recently published 1.3 mm detection of a dust disk around this third object and present for the first time a clear detection of a single peak {sup 12}CO (2–1) line, providing direct evidence for the simultaneous existence of a gas disk. We perform radiative transfer modeling of the third object in FW Tau and find that current observations are consistent with either a brown dwarf embedded in an edge-on disk or a planet embedded in a low inclination disk, which is externally irradiated by the binary companion. Further observations with ALMA, aiming for high SNR detections of non-contaminated gas lines, are required to conclusively unveil the nature of the third object in FW Tau.

79 ASTRONOMY AND ASTROPHYSICS↗

SU-F-J-159: Influence of the Elevated Posterior Position by Using the Customized Vacuum-Bag On the Abdominal MR Image Quality: A Quantitative Phantom Study

Purpose: Signal-to-noise ratio(SNR) of MR abdominal imaging in diagnostic radiology is maximized by minimizing the coil-to-patient distance. However, for radiotherapy applications, customized vacuum-bag is needed for abdominal immobilization at the cost of the increasing distance to the posterior spine coil. This sub-optimized coil setting for RT applications may compromise image quality, such as SNR and homogeneity, thus potentially affect tissue delineation. In this study, we quantitatively evaluate the effect of the vertical position change on SNR and image quality change using an ACR MR phantom. Methods: An ACR MR phantom was placed on the flat couch top. Images were acquired using an 18-channel body array coil and spine coil on a dedicated 1.5T MR-simulator. The scan was repeated three times with the ACR phantom elevated up to 7.5cm from the couch top, with a step size of 2.5cm. All images were acquired using standard ACR test sequence protocol of 2D spin-echo T1-weighted(TR/TE=500/200ms) and T2-weighted(TR/TE1/TE2=2000/20/80) sequences. For all scans, pre-scan normalization was turned on, and the distance between the phantom and the anterior 18-channel body array coil was kept constant. SNR was calculated using the slice with a large water-only region of the phantom. Percent intensity uniformity(PIU) and low contrast object detectability(LCD) were assessed by following ACR test guidelines. Results: The decrease in image SNR(from 335.8 to 169.3) and LCD(T1: from 31 to 19 spokes, T2: 26 to 16 spokes) were observed with increasing vertical distance. After elevating the phantom by 2.5cm(approximately the thickness of standard vacuum-bag), SNR change(from 335.8 to 275.5) and LCD(T1: 31 to 26 spokes, T2: 26 to 21 spokes) change were noted. However, similar PIU was obtained for all choices of vertical distance (T1: 94.5%–95.0%, T2: 94.4%–96.8%). Conclusion: After elevating the scan object, reduction in SNR level and contrast detectability but no change in image homogeneity was observed.

60 APPLIED LIFE SCIENCES↗

SU-E-J-211: Assessing the Consistency of the ViewRay 0.35 T MRI System

Purpose: ViewRay is a novel image-guided radiotherapy approach with an integrated 0.35 T MR unit and three Cobalt-60 heads. In order to continuously ensure the high resolution and high soft tissue contrast available in MR images, we quantified a multitude of relevant imaging parameters over a period of six months to establish its stability and imaging quality. In the assessment process, consideration is also given to the need to establish the number of tests required to have confidence in the performance of the system for radiation therapy planning applications. Methods: Daily, weekly, monthly and annual imaging tests were performed over a period of six months using standardized phantoms (a 24 cm diameter sphere and the ACR phantom) to quantify the performance of the system. In addition to the ACR and NEMA recommended tests, we also included element testing, spatial integrity, Eddy current and magnetic field inhomogeneity measurements. The ACR test is used for assessing the following parameters for T1 and T2: geometric accuracy, high contrast spatial resolution, slice thickness accuracy, slice position accuracy, percent signal ghosting, low contrast object detectability. It also includes percent image uniformity (PIU) for T1. The NEMA test is primarily designed to check SNR and PIU. Results: Over the period of six months, all the parameters were maintained within the recommendations provided in the ACR and NEMA standards. PIU and SNR were found to be sensitive to malfunctions in the components of the multileaf collimators. Details of the findings will be presented. Conclusion: The data suggests that ViewRay imaging system has functioned in a consistent and reliable manner. MR imaging from ViewRay 0.35T system complies with the ACR and NEMA recommended acceptance standards.

60 APPLIED LIFE SCIENCES↗

Synthetic Tracking on a Small Telescope

Synthetic tracking uses high speed (up to 10 Hz) low noise (<2e-) large format sensors ~16 Mpix along with a multi-vector shift/add algorithm that coadds multiple image frames to increase the signal to noise ratio (SNR) needed to detect (if present) multiple moving objects in the field of view (FOV). We published the application of synthetic tracking to look for asteroids in 2014 (Shao 2014), but recently have applied it more as well to Earth orbiting objects. We have begun testing the data processing graphical processing unit (GPU) array with a small telescope, a 28 cm Celestron RASA telescope and a low cost low noise 16 Mpix CMOS camera at a dark site in California. This system is now operational with a 2 sqdeg FOV and a limiting magnitude between ~16-17.5 stellar magnitudes (mag) depending on a number of observational parameters for short integration times. The instrument can be used to search for NEOs, where we use much longer integration times to get sensitivity ~ 20.5 mag (at new moon). Synthetic tracking provides significant improvements in both sensitivity and astrometric accuracy.

Turyshev, Slava G.↗

Multi-frame moving object detection system

A plurality of remote sensing images of a scene are received. A potential target object can be identified in one of the images, wherein the target object has a low signal-to-noise ratio (SNR). A candidate motion path of the target object can be generated based upon the images. A predicted position of the target object along the candidate motion path is determined for each of the remote sensing images. An image chip is extracted from each of the images, where each image chip is centered about the predicted position of the target object in its corresponding image. A sum image chip is generated based upon the image chips. An indication that the potential target object is an actual object in the images is output based upon a value of a center pixel of the sum image chip.

Ma, Tian J.↗

TYPE Ia SUPERNOVA REMNANT SHELL AT z = 3.5 SEEN IN THE THREE SIGHTLINES TOWARD THE GRAVITATIONALLY LENSED QSO B1422+231

Using the Subaru 8.2 m Telescope with the IRCS Echelle spectrograph, we obtained high-resolution (R = 10,000) near-infrared (1.01-1.38 {mu}m) spectra of images A and B of the gravitationally lensed QSO B1422+231 (z = 3.628) consisting of four known lensed images. We detected Mg II absorption lines at z = 3.54, which show a large variance of column densities ({approx}0.3 dex) and velocities ({approx}10 km s{sup -1}) between sightlines A and B with a projected separation of only 8.4h{sup -1}{sub 70} pc at that redshift. This is the smallest spatial structure of the high-z gas clouds ever detected after Rauch et al. found a 20 pc scale structure for the same z = 3.54 absorption system using optical spectra of images A and C. The observed systematic variances imply that the system is an expanding shell as originally suggested by Rauch et al. By combining the data for three sightlines, we managed to constrain the radius and expansion velocity of the shell ({approx}50-100 pc, 130 km s{sup -1}), concluding that the shell is truly a supernova remnant (SNR) rather than other types of shell objects, such as a giant H II region. We also detected strong Fe II absorption lines for this system, but with much broader Doppler width than that of {alpha}-element lines. We suggest that this Fe II absorption line originates in a localized Fe II-rich gas cloud that is not completely mixed with plowed ambient interstellar gas clouds showing other {alpha}-element low-ion absorption lines. Along with the Fe richness, we conclude that the SNR is produced by an SN Ia explosion.

79 ASTRONOMY AND ASTROPHYSICS↗

HFPK 334: An unusual supernova remnant in the Small Magellanic Cloud

We present new Australia Telescope Compact Array radio-continuum and XMM-Newton/Chandra X-ray Observatory observations of the unusual supernova remnant (SNR) HFPK 334 in the Small Magellanic Cloud (SMC). The remnant follows a shell-type morphology in the radio continuum and has a size of ∼20 pc at the SMC distance. The X-ray morphology is similar; however, we detect a prominent point source close to the center of the SNR exhibiting a spectrum with a best-fit power law with a photon index of Γ = 2.7 ± 0.5. This central point source is most likely a background object and cannot be directly associated with the remnant. The high temperature, nonequilibrium conditions in the diffuse region suggest that this gas has been recently shocked and points toward a younger SNR with an age of ≲ 1800 yr. With an average radio spectral index of α = –0.59 ± 0.09, we find that an equipartition magnetic field for the remnant is ∼90 μG, a value typical of younger SNRs in low-density environments. Also, we report the detection of scattered radio polarization across the remnant at 20 cm, with a peak fractional polarization level of 25% ± 5%.

79 ASTRONOMY AND ASTROPHYSICS↗

Molecular Environments of Three Large Supernova Remnants in the Third Galactic Quadrant: G205.5+0.5, G206.9+2.3, and G213.0–0.6

We present CO observations toward three large supernova remnants (SNRs) in the third Galactic quadrant using the Purple Mountain Observatory Delingha 13.7 m millimeter-wavelength telescope. The observations are part of the high-resolution CO survey of the Galactic plane between Galactic longitudes l=−10{sup ∘} to 250{sup ∘} and latitudes b=−5{sup ∘} to 5{sup ∘}. CO emission was detected toward the three SNRs: G205.5+0.5 (Monoceros Nebula), G206.9+2.3 (PKS 0646+06), and G213.0–0.6. Both SNRs G205.5+0.5 and G213.0–0.6 exhibit the morphological agreement (or spatial correspondences) between the remnant and the surrounding molecular clouds (MCs), as well as kinematic signatures of shock perturbation in the molecular gas. We confirm that the two SNRs are physically associated with their ambient MCs and the shock of SNRs is interacting with the dense, clumpy molecular gas. SNR G206.9+2.3, which is close to the northeastern edge of the Monoceros Nebula, displays the spatial coincidence with molecular partial shell structures at V {sub LSR} ∼ 15 km s{sup −1}. While no significant line broadening has been detected within or near the remnant, the strong morphological correspondence between the SNR and the molecular cavity implies that SNR G206.9+2.3 is probably associated with the CO gas and is evolving in the low-density environment. The physical features of individual SNRs, together with the relationship between SNRs and their nearby objects, are also discussed.

79 ASTRONOMY AND ASTROPHYSICS↗

THE NATURE OF THE STRONG 24 {mu}m SPITZER SOURCE J222557+601148: NOT A YOUNG GALACTIC SUPERNOVA REMNANT

The nebula J222557+601148, tentatively identified by Morris et al. as a young Galactic supernova remnant (SNR) from Spitzer Galactic First Look Survey images and a follow-up mid-infrared spectrum, is unlikely to be an SNR remnant based on H{alpha}, [O III], [S II] images, and low-dispersion optical spectra. The object is seen in H{alpha} and [O III] {lambda}5007 images as a faint, roughly circular ring nebula with dimensions matching that seen in 24 {mu}m Spitzer images. Low-dispersion optical spectra show it to have narrow H{alpha} and [N II] {lambda}{lambda}6548,6583 line emissions with no evidence of broad or high-velocity (v {>=} 300 km s{sup -1}) line emissions. The absence of any high-velocity optical features, the presence of relatively strong [N II] emissions, the lack of detected [S II] emission which would indicate the presence of shock-heated gas, plus no coincident X-ray or nonthermal radio emissions indicate that the nebula is unlikely to be an SNR, young or old. Instead, it is likely a faint, high-excitation planetary nebula (PN) as its elliptical morphology would suggest, lying at a distance {approx}2-3 kpc with unusual but not extraordinary mid-IR colors and spectra. We have identified an m{sub r'} = 22.4 {+-} 0.2 star as a PN central star candidate.

79 ASTRONOMY AND ASTROPHYSICS↗

EPIC Study of Two Enigmatic Sources: The Mouse and SNR 359.1-0.5

The objective of the original proposal was to observe the Mouse pulsar wind nebula (associated with PSR J1744-2958) and the nearby supernova remnant G359.1-0.5, where the pulsar was probably born, with the XMM-Newton observatory to study the properties of these objects. SNR G359.1-0.5 was accepted as a Category C target and has not been observed. The Mouse was observed on April 27,2003 for 52 ks. The image analysis has shown that the Mouse is extended in the East-West direction, possibly along the direction of the pulsar's proper motion. The spectrum of this pulsar wind nebula can be described as an absorbed power law with the photon index GAMMA = 1.9 plus or minus 0.1, effective hydrogen column density n(sub H) = (2.6 plus or minus 0.1) x 10(exp 22) per square centimeter, and flux F = 1.8 x 10(exp -11) erg per square centimeter per second in the 1-10 keV energy range. Based on the n(sub H) value, the distance to the source is about 5 kpc, which results in the luminosity 3.7 x 10(exp 34) erg per second. We conclude that PSR J1744-2958 and the Mouse are not physically associated with G359.1-0.5, which lies at a larger distance. In addition to the Mouse, we also detected two Low-Mass X-ray Binaries, SLX 1744-299 and 1744-300, in the EPIC MOS and PN fields of view. The latter of these objects showed a Type I X-ray burst during our observation, with a rise time of 5 s and decay time of 60 s. A very strong pileup during the burst made the analysis of the burst properties unreliable. The spectral analysis of the persistent radiation from SLX 1744-299 and 1744-300 yields the hydrogen column densities of 3.2 plus or minus 0.1 and (3.6 plus or minus 0.2) x 10(exp 22) per square centimeter, respectively, which suggests that the sources are close to the Galactic center (d = 8-9 kpc). The spectra can be reasonably well fitted with a blackbody plus thin disk model, with the blackbody temperatures of 1.7 plus or minus 0.2 and 1.8 plus or minus 0.2 keV, respectively.

Pavlov, George↗

The effect of averaging adjacent planes for artifact reduction in matrix inversion tomosynthesis

Purpose: Matrix inversion tomosynthesis (MITS) uses linear systems theory and knowledge of the imaging geometry to remove tomographic blur that is present in conventional backprojection tomosynthesis reconstructions, leaving in-plane detail rendered clearly. The use of partial-pixel interpolation during the backprojection process introduces imprecision in the MITS modeling of tomographic blur, and creates low-contrast artifacts in some MITS planes. This paper examines the use of MITS slabs, created by averaging several adjacent MITS planes, as a method for suppressing partial-pixel artifacts. Methods: Human chest tomosynthesis projection data, acquired as part of an IRB-approved pilot study, were used to generate MITS planes, three-plane MITS slabs (MITSa3), five-plane MITS slabs (MITSa5), and seven-plane MITS slabs (MITSa7). These were qualitatively examined for partial-pixel artifacts and the visibility of normal and abnormal anatomy. Additionally, small (5 mm) subtle pulmonary nodules were simulated and digitally superimposed upon human chest tomosynthesis projection images, and their visibility was qualitatively assessed in the different reconstruction techniques. Simulated images of a thin wire were used to generate modulation transfer function (MTF) and slice-sensitivity profile curves for the different MITS and MITS slab techniques, and these were examined for indications of partial-pixel artifacts and frequency response uniformity. Finally, mean-subtracted, exposure-normalized noise power spectra (ENNPS) estimates were computed and compared for MITS and MITS slab reconstructions, generated from 10 sets of tomosynthesis projection data of an acrylic slab. The simulated in-plane MTF response of each technique was also combined with the square root of the ENNPS estimate to yield stochastic signal-to-noise ratio (SNR) information about the different reconstruction techniques. Results: For scan angles of 20 Degree-Sign and 5 mm plane separation, seven MITS planes must be averaged to sufficiently remove partial-pixel artifacts. MITSa7 does appear to subtly reduce the contrast of high-frequency 'edge' information, but the removal of partial-pixel artifacts makes the appearance of low-contrast, fine-detail anatomy even more conspicuous in MITSa7 slices. MITSa7 also appears to render simulated subtle 5 mm pulmonary nodules with greater visibility than MITS alone, in both the open lung and regions overlying the mediastinum. Finally, the MITSa7 technique reduces stochastic image variance, though the in-plane stochastic SNR (for very thin objects which do not span multiple MITS planes) is only improved at spatial frequencies between 0.05 and 0.20 cycles/mm. Conclusions: The MITSa7 method is an improvement over traditional single-plane MITS for thoracic imaging and the pulmonary nodule detection task, and thus the authors plan to use the MITSa7 approach for all future MITS research at the authors' institution.

60 APPLIED LIFE SCIENCES↗

A mathematical model of extremely low frequency ocean induced electromagnetic noise

Magnetic Anomaly Detection (MAD) system uses the principle that ferromagnetic objects disturb the magnetic lines of force of the earth. These lines of force are able to pass through both water and air in similar manners. A MAD system, usually mounted on an aerial vehicle, is thus often employed to confirm the detection and accomplish localization of large ferromagnetic objects submerged in a sea-water environment. However, the total magnetic signal encountered by a MAD system includes contributions from a myriad of low to Extremely Low Frequency (ELF) sources. The goal of the MAD system is to detect small anomaly signals in the midst of these low-frequency interfering signals. Both the Range of Detection (R{sub d}) and the Probability of Detection (P{sub d}) are limited by the ratio of anomaly signal strength to the interfering magnetic noise. In this paper, we report a generic mathematical model to estimate the signal-to-noise ratio or SNR. Since time-variant electro-magnetic signals are affected by conduction losses due to sea-water conductivity and the presence of air-water interface, we employ the general formulation of dipole induced electromagnetic field propagation in stratified media [1]. As a first step we employ a volumetric distribution of isolated elementary magnetic dipoles, each having its own dipole strength and orientation, to estimate the magnetic noise observed by a MAD system. Numerical results are presented for a few realizations out of an ensemble of possible realizations of elementary dipole source distributions.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Beyond Contrast Transfer: Spectral SNR as a Finite-Dose Metric for STEM Phase Retrieval

The contrast transfer function (CTF) is widely used to evaluate phase retrieval methods in scanning transmission electron microscopy (STEM), including center-of-mass imaging, parallax imaging, direct ptychography, and iterative ptychography. However, the CTF reflects only the maximum usable signal, neglecting the effects of finite electron fluence and the Poisson-limited nature of detection. As a result, it can significantly overestimate practical performance, especially in low-dose regimes. Here, we employ the spectral signal-to-noise ratio (SSNR), as a finite-dose statistical framework to evaluate the recoverable signal as a function of spatial frequency. Using numerical reconstructions of white-noise objects, we show that center-of-mass, parallax, and direct ptychography exhibit dose-independent SSNRs, with close-form analytic expressions. In contrast, iterative ptychography exhibits a surprising dose dependence: at low fluence, its SSNR converges to that of direct ptychography; at high fluence, it saturates at a value consistent with the maximum detective quantum efficiency predicted by recent quantum Fisher information bounds. The results highlight the limitations of CTF-based evaluation and motivate SSNR as a more accurate, finite-dose metric for assessing STEM phase retrieval methods.

STEM phase retrieval↗

Spectrally Resolved Synthetic Aperture Imaging Interferometer: Final Report of NASA Innovative Advanced Concepts - Phase 1 Study

The Spectrally Resolved Synthetic-Aperture Imaging Interferometer (SRSAII) is a system proposed to provide high-resolution and high-sensitivity measurements of astronomical objects. SRSAII uses long baseline interferometric methods to achieve the resolution and low-noise, high time-precision detection to achieve the sensitivity. The primary goal of the SRSAII study was to lay out a framework for using new optical physics technologies to directly resolve, both spatially and spectrally, the disk of an exoplanet. In addition to the ambitious goal of directly resolving an exoplanet, the SRSAII team also sought to identify science opportunities achievable with intermediate system configurations which may offer resolution significantly higher than the current state of the art, but insufficient for direct resolution of an exoplanetary disk. An operational SRSAII system can function with essentially arbitrarily large baselines, achieving correspondingly high angular resolution. The primary limitation occurs in the system sensitivity, which became the major technical focus for study. In this report, we compare the predicted performance (sensitivity in SNR (Signal-to-Noise Ratio) along with angular resolution) of three interferometric techniques: direct detection (also known as homodyne interferometry), multi-channel intensity interferometry (using the Hanbury Brown and Twiss effect), and multi-channel heterodyne interferometry (using an optical frequency comb as a local oscillator). Additionally, quantum-assisted interferometry is also explored as a prospective enhancement of established methods. This report presents a survey of the technologies that enable the SRSAII techniques - optical frequency combs, single photon detectors, and photonic integrated circuits. These technologies are the basis of methods critical to SRSAII's success: precision timing, length and frequency metrology, sensitive photodetection, fine-scale wavelength filtering, and dense multi-channel operation. Lastly, we give some notional performance metrics and propose some possible experimental observations.

High Resolultion↗

Background based Gaussian mixture model lesion segmentation in PET

Purpose: Quantitative {sup 18}F-fluorodeoxyglucose positron emission tomography is limited by the uncertainty in lesion delineation due to poor SNR, low resolution, and partial volume effects, subsequently impacting oncological assessment, treatment planning, and follow-up. The present work develops and validates a segmentation algorithm based on statistical clustering. The introduction of constraints based on background features and contiguity priors is expected to improve robustness vs clinical image characteristics such as lesion dimension, noise, and contrast level. Methods: An eight-class Gaussian mixture model (GMM) clustering algorithm was modified by constraining the mean and variance parameters of four background classes according to the previous analysis of a lesion-free background volume of interest (background modeling). Hence, expectation maximization operated only on the four classes dedicated to lesion detection. To favor the segmentation of connected objects, a further variant was introduced by inserting priors relevant to the classification of neighbors. The algorithm was applied to simulated datasets and acquired phantom data. Feasibility and robustness toward initialization were assessed on a clinical dataset manually contoured by two expert clinicians. Comparisons were performed with respect to a standard eight-class GMM algorithm and to four different state-of-the-art methods in terms of volume error (VE), Dice index, classification error (CE), and Hausdorff distance (HD). Results: The proposed GMM segmentation with background modeling outperformed standard GMM and all the other tested methods. Medians of accuracy indexes were VE <3%, Dice >0.88, CE <0.25, and HD <1.2 in simulations; VE <23%, Dice >0.74, CE <0.43, and HD <1.77 in phantom data. Robustness toward image statistic changes (±15%) was shown by the low index changes: <26% for VE, <17% for Dice, and <15% for CE. Finally, robustness toward the user-dependent volume initialization was demonstrated. The inclusion of the spatial prior improved segmentation accuracy only for lesions surrounded by heterogeneous background: in the relevant simulation subset, the median VE significantly decreased from 13% to 7%. Results on clinical data were found in accordance with simulations, with absolute VE <7%, Dice >0.85, CE <0.30, and HD <0.81. Conclusions: The sole introduction of constraints based on background modeling outperformed standard GMM and the other tested algorithms. Insertion of a spatial prior improved the accuracy for realistic cases of objects in heterogeneous backgrounds. Moreover, robustness against initialization supports the applicability in a clinical setting. In conclusion, application-driven constraints can generally improve the capabilities of GMM and statistical clustering algorithms.

60 APPLIED LIFE SCIENCES↗