Howard C. Gifford

A comparison of human and model observers in multislice LROC studies

Model and human observers have been compared in a series of localization receiver operating characteristic (LROC) studies involving single-slice and multislice image displays. The task was detection of Ga-avid lymphomas within single photon emission computed tomography (SPECT)-reconstructed transverse slices of a mathematical phantom, and the studies involved four reconstruction strategies: the filtered-backprojection (FBP) and ordered-subset expectation-maximization (OSEM) algorithms with two- and three-dimensional postreconstruction filtering. The human-observer data was drawn from studies performed by Wells et al. (2000), while multiclass versions of the nonprewhitening (NPW), channelized nonprewhitening (CNPW), and channelized Hotelling (CH) model observers, each capable of performing the tumor search task, were applied. The channelized observers were evaluated with multiple square-channel models and both with and without internal noise. For the multislice studies, two different capacities for integrating the slice information were also tested. The CH observer gave good quantitative agreement with the human data from both image-display studies when the internal-noise model was used. The CNPW observer performed similarly with the iterative strategies. Wells et al. had shown that human observers are imperfect integrators of multislice information, and this is characterized as increased internal noise with the model observers.

Motion correction for cardiac SPECT using a RBI-ML partial-reconstruction approach

In single photon computed tomography, patient motion can significantly affect image quality. Methods to correct for patient motion rely on available information about the motion or stable algorithms have to be developed to detect and describe object motion. In this work we introduce a reconstruction method that corrects for rigid body motion and we investigate a method to exploit the motion information that might be hidden in the projection data and possibly modify motion descriptions that were retrieved by external motion tracking devices. The search method was based on reconstructing only parts of the projection angles using an RBI-ML partial-reconstruction approach (PRA). It tests a motion description with only one forward projection per detector head instead of a full reconstruction using all projections. A figure of merit based on the per angle likelihood function in projection space was introduced. With simulated MCAT data we could show that the PRA was able to identify the optimum 3D rigid body motion correction with an error of 1 pixel in each cartesian direction. Future work will address 3D translations, multiple motions, detector resolution compensation and will also concentrate on the improvement of the figure of merit.

An optimization of reconstruction parameters and investigation into the impact of photon scatter in /sup 67/Ga SPECT

During /sup 67/Ga citrate SPECT imaging, photon downscatter will occur from higher energy photons into lower energy acquisition windows thus possibly adversely affecting reconstructed image quality. With these additional scattered photons present in projection data, the effect of using more complex reconstruction strategies such as three-dimensional detector response compensation (3-D DRC) and attenuation correction (AC) is unclear. Using a combination of numerical channelized hotelling observers (CHO) and human localization receiver operating characteristics (LROC) studies it has been found that maximum lesion detectability occurs when projection data is reconstructed using two iterations of the rescaled block iterative (RBI) algorithm with both 3-D DRC and patient specific AC, followed by a postreconstruction low-pass 3-D Gaussian filtering with a FWHM of /spl ap/1 cm. These parameters deviate from optimal reconstruction parameters for primary photon-only projection data which finds that maximum lesion detection occurs after 4 RBI iterations with the same 3-D DRC, AC, and postfiltering. As expected, it is also observed that decreased tumor detectability results when scattered photons are present in /sup 67/Ga projection data compared to primary photon only reconstructions (reducing maximum A/sub L/ from 0.79 to 0.58). This decrease has been found to be statistically significant in both human LROC and numerical observer studies, suggesting the need for scatter compensation during /sup 67/Ga citrate imaging.

Optimization of regularization of attenuation and scatter corrected /sup 99 m/Tc cardiac SPECT studies for defect detection using hybrid images

Through means of an ROC study, the authors optimize the iteration number and 3-D Gaussian post-filtering of /sup 99 m/Tc cardiac emission OSEM reconstructions that implement corrections for both attenuation and scatter. Hybrid images were used for this optimization wherein perfusion defects were added artificially to clinical patient studies that were read as being normally perfused. The test conditions included 3 different iteration numbers of OSEM (1, 5 and 10), followed by 3-D Gaussian low-pass filtering at each iteration level. The level of Gaussian low-pass filtering was varied using standard deviations (/spl sigma/) of 0.6, 0.75, 1 and 1.25 pixels, in addition to a case where no post-filtering was applied. Four observers read 80 images for each of the 15 test conditions being investigated, providing confidence ratings as to the presence or absence of perfusion defects. Results indicate a slowly varying trend between very little filtering and quite heavy levels of smoothing with a gentle plateau for post-filters in the range of /spl sigma/=0.6 to 1 pixel. No significant improvement in detection accuracy was observed with increasing iteration number as long as the reconstructions are post-filtered with /spl sigma/ in the range of 0.6 to 1 pixel, suggesting that 1 complete iteration of OSEM should suffice.

Visual-search observers for assessing tomographic x-ray image quality.

PURPOSE Mathematical model observers commonly used for diagnostic image-quality assessments in x-ray imaging research are generally constrained to relatively simple detection tasks due to their need for statistical prior information. Visual-search (VS) model observers that employ morphological features in sequential search and analysis stages have less need for such information and fewer task constraints. The authors compared four VS observers against human observers and an existing scanning model observer in a pilot study that quantified how mass detection and localization in simulated digital breast tomosynthesis (DBT) can be affected by the number P of acquired projections. METHODS Digital breast phantoms with embedded spherical masses provided single-target cases for a localization receiver operating characteristic (LROC) study. DBT projection sets based on an acquisition arc of 60° were generated for values of P between 3 and 51. DBT volumes were reconstructed using filtered backprojection with a constant 3D Butterworth postfilter; extracted 2D slices were used as test images. Three imaging physicists participated as observers. A scanning channelized nonprewhitening (CNPW) observer had knowledge of the mean lesion-absent images. The VS observers computed an initial single-feature search statistic that identified candidate locations as local maxima of either a template matched-filter (MF) image or a gradient-template MF (GMF) image. Search inefficiencies that modified the statistic were also considered. Subsequent VS candidate analyses were carried out with (i) the CNPW statistical discriminant and (ii) the discriminant computed from GMF training images. These location-invariant discriminants did not utilize covariance information. All observers read 36 training images and 108 study images per P value. Performance was scored in terms of area under the LROC curve. RESULTS Average human-observer performance was stable for P between 7 and 35. In the absence of search inefficiencies, the VS models based on the GMF analysis provided the best correlation (Pearson ρ ≥ 0.62) with the human results. The CNPW-based VS observers deviated from the humans primarily at lower values of P. In this limited study, search inefficiencies allowed for good quantitative agreement with the humans for most of the VS observers. CONCLUSIONS The computationally efficient training requirements for the VS observer are suitable for high-resolution imaging, indicating that the observer framework has the potential to overcome important task limitations of current model observers for x-ray applications.

Accounting for anatomical noise in search-capable model observers for planar nuclear imaging

Abstract. Model observers intended to predict the diagnostic performance of human observers should account for the effects of both quantum and anatomical noise. We compared the abilities of several visual-search (VS) and scanning Hotelling-type models to account for anatomical noise in a localization receiver operating characteristic (LROC) study involving simulated nuclear medicine images. Our VS observer invoked a two-stage process of search and analysis. The images featured lesions in the prostate and pelvic lymph nodes. Lesion contrast and the geometric resolution and sensitivity of the imaging collimator were the study variables. A set of anthropomorphic mathematical phantoms was imaged with an analytic projector based on eight parallel-hole collimators with different sensitivity and resolution properties. The LROC study was conducted with human observers and the channelized nonprewhitening, channelized Hotelling (CH) and VS model observers. The CH observer was applied in a “background-known-statistically” protocol while the VS observer performed a quasi-background-known-exactly task. Both of these models were applied with and without internal noise in the decision variables. A perceptual search threshold was also tested with the VS observer. The model observers without inefficiencies failed to mimic the average performance trend for the humans. The CH and VS observers with internal noise matched the humans primarily at low collimator sensitivities. With both internal noise and the search threshold, the VS observer attained quantitative agreement with the human observers. Computational efficiency is an important advantage of the VS observer.

The impact of noisy and misaligned attenuation maps on human-observer performance at lesion detection in SPECT

We have demonstrated an improvement due to attenuation correction (AC) at the task of lesion detection in thoracic SPECT images. However, increased noise in the transmission data due to aging sources or very large patients, and misregistration of the emission and transmission maps, can reduce the accuracy of the AC and may result in a loss of lesion detectability. We investigated the impact of noise in and misregistration of transmission data, on the detection of simulated Ga-67 thoracic lesions. Human-observer localization-receiver-operating-characteristic (LROC) methodology was used to assess performance. Both emission and transmission data were simulated using the MCAT computer phantom. Emission data were reconstructed using OSEM incorporating AC and detector resolution compensation. Clinical noise levels were used in the emission data. The transmission-data noise levels ranged from zero (noise-free) to 32 times the measured clinical levels. Transaxial misregistrations of 0.32, 0.63, and 1.27 cm between emission and transmission data were also examined. Three different algorithms were considered for creating the attenuation maps: filtered backprojection (FBP), unbounded maximum-likelihood (ML), and block-iterative transmission AB (BITAB). Results indicate that a 16-fold increase in the noise was required to eliminate the benefit afforded by AC, when using FBP or ML to reconstruct the attenuation maps. When using BITAB, no significant loss in performance was observed for a 32-fold increase in noise. Misregistration errors are also a concern as even small errors here reduce the performance gains of AC.

Assessment of prostate cancer detection with a visual-search human model observer

Early staging of prostate cancer (PC) is a significant challenge, in part because of the small tumor sizes in- volved. Our long-term goal is to determine realistic diagnostic task performance benchmarks for standard PC imaging with single photon emission computed tomography (SPECT). This paper reports on a localization receiver operator characteristic (LROC) validation study comparing human and model observers. The study made use of a digital anthropomorphic phantom and one-cm tumors within the prostate and pelvic lymph nodes. Uptake values were consistent with data obtained from clinical In-111 ProstaScint scans. The SPECT simulation modeled a parallel-hole imaging geometry with medium-energy collimators. Nonuniform attenua- tion and distance-dependent detector response were accounted for both in the imaging and the ordered-subset expectation-maximization (OSEM) iterative reconstruction. The observer study made use of 2D slices extracted from reconstructed volumes. All observers were informed about the prostate and nodal locations in an image. Iteration number and the level of postreconstruction smoothing were study parameters. The results show that a visual-search (VS) model observer correlates better with the average detection performance of human observers than does a scanning channelized nonprewhitening (CNPW) model observer.

Towards Visual-Search Model Observers for Mass Detection in Breast Tomosynthesis.

We are investigating human-observer models that perform clinically realistic detection and localization tasks as a means of making reliable assessments of digital breast tomosynthesis images. The channelized non-prewhitening (CNPW) observer uses the background known exactly task for localization and detection. Visual-search observer models attempt to replicate the search patterns of trained radiologists. The visual-search observer described in this paper utilizes a two-phase approach, with an initial holistic search followed by directed analysis and decision making. Gradient template matching is used for the holistic search, and the CNPW observer is used for analysis and decision making. Spherical masses were embedded into anthropomorphic breast phantoms, and simulated projections were made using ray-tracing and a serial cascade model. A localization ROC study was performed on these images using the visual-search model observer and the CNPW observer. Observer performance from the two computer observers was compared to human observer performance. The visual-search observer was able to produce area under the LROC curve values similar to those from human observers; however, more research is needed to increase the robustness of the algorithm.

Assessment of scatter compensation strategies for /sup 67/Ga tumor SPECT using numerical observers and human LROC studies

Ga-67 citrate SPECT imaging is often used for oncological studies in order to diagnose or stage patient lymphomas. Because the decay of Ga-67 involves multiple emission energies, it is possible that many down-scattered photons will be present in photopeak acquisition data. We have previously shown through human observer LROC studies, that the inclusion of these scattered photons significantly degrades lesion detectability in simulations. We have investigated the use of six different scatter compensation methods representing different strategies. These consist of i) perfect scatter rejection, ii) no scatter compensation, iii) ideal scatter compensation, iv) triple energy window estimation, v) effective scatter source estimation, and vi) post-reconstruction scatter subtraction. Each method has first been optimized using a channelized hotelling numerical observer, then ranked through the use of a human LROC study and by using a newly devised LROC numerical observer. Both human LROC and LROC numerical observer results indicate that both TEW and ESSE scatter compensation methods are able to improve lesion detectability over no compensation, but fail to achieve similar detectability to using perfect scatter rejection. Excellent agreement between the LROC numerical observer and human LROC studies indicate that the LROC observer may be good predictor of human performance in Ga-67 SPECT.