Alexandre Ba

Anthropomorphic model observer performance in three-dimensional detection task for low-contrast computed tomography

Abstract. X-ray medical imaging is increasingly becoming three-dimensional (3-D). The dose to the population and its management are of special concern in computed tomography (CT). Task-based methods with model observers to assess the dose-image quality trade-off are promising tools, but they still need to be validated for real volumetric images. The purpose of the present work is to evaluate anthropomorphic model observers in 3-D detection tasks for low-contrast CT images. We scanned a low-contrast phantom containing four types of signals at three dose levels and used two reconstruction algorithms. We implemented a multislice model observer based on the channelized Hotelling observer (msCHO) with anthropomorphic channels and investigated different internal noise methods. We found a good correlation for all tested model observers. These results suggest that the msCHO can be used as a relevant task-based method to evaluate low-contrast detection for CT and optimize scan protocols to lower dose in an efficient way.

PATIENT EXPOSURE OPTIMISATION THROUGH TASK-BASED ASSESSMENT OF A NEW MODEL-BASED ITERATIVE RECONSTRUCTION TECHNIQUE

The goal of the present work was to report and investigate the performances of a new iterative reconstruction algorithm, using a model observer. For that, a dedicated low-contrast phantom containing different targets was scanned at four volume computed tomography dose index (CTDIvol) levels on a Siemens SOMATOM Force computed tomography (CT). The acquired images were reconstructed using the ADMIRE algorithm and were then assessed by three human observers who performed alternative forced choice experiments. Next, a channelised hotelling observer model was applied on the same set of images. The comparison between the two was performed using the percentage correct as a figure of merit. The results indicated a strong agreement between human and model observer as well as an improvement in the low-contrast detection when switching from an ADMIRE strength of 1–3. Good results were also observed even in situations where the target was hard to detect, suggesting that patient dose could be further reduced and optimised.

OBJECTIVE TASK-BASED ASSESSMENT OF LOW-CONTRAST DETECTABILITY IN ITERATIVE RECONSTRUCTION

Evaluating image quality by using receiver operating characteristic studies is time consuming and difficult to implement. This work assesses a new iterative algorithm using a channelised Hotelling observer (CHO). For this purpose, an anthropomorphic abdomen phantom with spheres of various sizes and contrasts was scanned at 3 volume computed tomography dose index (CTDIvol) levels on a GE Revolution CT. Images were reconstructed using the iterative reconstruction method adaptive statistical iterative reconstruction-V (ASIR-V) at ASIR-V 0, 50 and 70 % and assessed by applying a CHO with dense difference of Gaussian and internal noise. Both CHO and human observers (HO) were compared based on a four-alternative forced-choice experiment, using the percentage correct as a figure of merit. The results showed accordance between CHO and HO. Moreover, an improvement in the low-contrast detection was observed when switching from ASIR-V 0 to 50 %. The results underpin the finding that ASIR-V allows dose reduction.

Assessment of low contrast detection in CT using model observers: Developing a clinically-relevant tool for characterising adaptive statistical and model-based iterative reconstruction

PURPOSE This study aims to assess CT image quality in a way that would meet specific requirements of clinical practice. Physics metrics like Fourier transform derived metrics were traditionally employed for that. However, assessment methods through a detection task have also developed quite extensively lately, and we chose here to rely on this modality for image quality assessment. Our goal was to develop a tool adapted for a fast and reliable CT image quality assessment in order to pave the way for new CT benchmarking techniques in a clinical context. Additionally, we also used this method to estimate the benefits brought by some IR algorithms. MATERIALS AND METHODS A modified QRM chest phantom containing spheres of 5 and 8mm at contrast levels of 10 and 20HU at 120kVp was used. Images of the phantom were acquired at CTDIvol of 0.8, 3.6, 8.2 and 14.5mGy, before being reconstructed using FBP, ASIR 40 and MBIR on a GE HD 750 CT scanner. They were then assessed by eight human observers undergoing a 4-AFC test. After that, these data were compared with the results obtained from two different model observers (NPWE and CHO with DDoG channels). The study investigated the effects of the acquisition conditions as well as reconstruction methods. RESULTS NPWE and CHO models both gave coherent results and approximated human observer results well. Moreover, the reconstruction technique used to retrieve the images had a clear impact on the PC values. Both models suggest that switching from FBP to ASIR 40 and particularly to MBIR produces an increase of the low contrast detection, provided a minimum level of exposure is reached. CONCLUSION Our work shows that both CHO with DDoG channels and NPWE models both approximate the trend of humans performing a detection task. Both models also suggest that the use of MBIR goes along with an increase of the PCs, indicating that further dose reduction is still possible when using those techniques. Eventually, the CHO model associated to the protocol we described in this study happened to be an efficient way to assess CT images in a clinical environment. In the future, this simple method could represent a sound basis to benchmark clinical practice and CT devices.

Objective comparison of high‐contrast spatial resolution and low‐contrast detectability for various clinical protocols on multiple CT scanners

Purpose We sought to compare objectively computed tomography (CT) scanner performance for three clinically relevant protocols using a task‐based image quality assessment method in order to assess the potential for radiation dose reduction. Methods Four CT scanners released between 2003 and 2007 by different manufacturers were compared with four CT scanners released between 2012 and 2014 by the same manufacturers using ideal linear model observers (MO): prewhitening (PW) MO and channelized Hotelling (CHO) MO with Laguerre‐Gauss channels for high‐contrast spatial resolution and low‐contrast detectability (LCD) performance, respectively. High‐contrast spatial resolution was assessed using a custom‐made phantom that enabled the computation of the target transfer function (TTF) and noise power spectrum (NPS). Low‐contrast detectability was assessed using a commercially available anthropomorphic abdominal phantom providing equivalent diameters of 24, 29.6, and 34.6 cm. Three protocols were reviewed: a head (trauma) and an abdominal (urinary stones) protocol were applied to assess high‐contrast spatial resolution performance; and another abdominal (focal liver lesions) protocol was applied for LCD. The liver protocol was tested using fixed and modulated tube currents. The PW MO was proposed for assessing high‐contrast detectability performance of the various CT scanners. Results Compared with older generation CT scanners, three newer systems displayed significant improvements in high‐contrast detectability over that of their predecessors. A fourth, newer system had lower performance. The CHO MO was appropriate for assessing LCD performance and revealed that an excellent level of image quality could be obtained with newer scanners at significantly lower dose levels. Conclusions This study shows that MO can objectively benchmark CT scanners using a task‐based image quality method, thus helping to estimate the potential for further dose reductions offered by the latest systems. Such an approach may be useful for adequately and quantitatively comparing clinically relevant image quality among various scanners.

BENCHMARKING OF CT FOR PATIENT EXPOSURE OPTIMISATION

Patient dose optimisation in computed tomography (CT) should be done using clinically relevant tasks when dealing with image quality assessments. In the present work, low-contrast detectability for an average patient morphology was assessed on 56 CT units, using a model observer applied on images acquired with two specific protocols of an anthropomorphic phantom containing spheres. Images were assessed using the channelised Hotelling observer (CHO) with dense difference of Gaussian channels. The results were computed by performing receiver operating characteristics analysis (ROC) and using the area under the ROC curve (AUC) as a figure of merit. The results showed a small disparity at a volume computed tomography dose index (CTDIvol) of 15 mGy depending on the CT units for the chosen image quality criterion. For 8-mm targets, AUCs were 0.999 ± 0.018 at 20 Hounsfield units (HU) and 0.927 ± 0.054 at 10 HU. For 5-mm targets, AUCs were 0.947 ± 0.059 and 0.702 ± 0.068 at 20 and 10 HU, respectively. The robustness of the CHO opens the way for CT protocol benchmarking and optimisation processes.

Inter‐laboratory comparison of channelized hotelling observer computation

Purpose The task‐based assessment of image quality using model observers is increasingly used for the assessment of different imaging modalities. However, the performance computation of model observers needs standardization as well as a well‐established trust in its implementation methodology and uncertainty estimation. The purpose of this work was to determine the degree of equivalence of the channelized Hotelling observer performance and uncertainty estimation using an intercomparison exercise. Materials and Methods Image samples to estimate model observer performance for detection tasks were generated from two‐dimensional CT image slices of a uniform water phantom. A common set of images was sent to participating laboratories to perform and document the following tasks: (a) estimate the detectability index of a well‐defined CHO and its uncertainty in three conditions involving different sized targets all at the same dose, and (b) apply this CHO to an image set where ground truth was unknown to participants (lower image dose). In addition, and on an optional basis, we asked the participating laboratories to (c) estimate the performance of real human observers from a psychophysical experiment of their choice. Each of the 13 participating laboratories was confidentially assigned a participant number and image sets could be downloaded through a secure server. Results were distributed with each participant recognizable by its number and then each laboratory was able to modify their results with justification as model observer calculation are not yet a routine and potentially error prone. Results Detectability index increased with signal size for all participants and was very consistent for 6 mm sized target while showing higher variability for 8 and 10 mm sized target. There was one order of magnitude between the lowest and the largest uncertainty estimation. Conclusions This intercomparison helped define the state of the art of model observer performance computation and with thirteen participants, reflects openness and trust within the medical imaging community. The performance of a CHO with explicitly defined channels and a relatively large number of test images was consistently estimated by all participants. In contrast, the paper demonstrates that there is no agreement on estimating the variance of detectability in the training and testing setting.

A resampling comparison of CHO's detectability index bias and uncertainty

Model observers have gained popularity as a surrogate approach for image quality assessment, they are often used for the optimization of the reconstruction algorithm. The most widespread model observer is the channelized Hotelling observer (CHO) that allows measuring the image quality by calculating the detectability index (or associated area under receiver operating characteristic curve). In this work we have chosen to explore different resampling methods used to estimate the CHO performance and uncertainty. In this paper, using data from the inter-laboratory comparison of the computation of CHO model observer study, we established a simulation framework to fully evaluate different resampling methods, namely, leave-one out and bootstrapping with replacement to estimate the CHO’s detectability index bias and uncertainty. For this particular study, we focus our experiments on datasets with a few data samples, 200 normal and 200 abnormal images.

Task-based quantification of image quality using a model observer in abdominal CT: a multicentre study

ObjectiveWe investigated the variability in diagnostic information inherent in computed tomography (CT) images acquired at 68 different CT units, with the selected acquisition protocols aiming to answer the same clinical question.MethodsAn anthropomorphic abdominal phantom with two optional rings was scanned on 68 CT systems from 62 centres using the local clinical acquisition parameters of the portal venous phase for the detection of focal liver lesions. Low-contrast detectability (LCD) was assessed objectively with channelised Hotelling observer (CHO) using the receiver operating characteristic (ROC) paradigm. For each lesion size, the area under the ROC curve (AUC) was calculated and considered as a figure of merit. The volume computed tomography dose index (CTDIvol) was used to indicate radiation dose exposure.ResultsThe median CTDIvol used was 5.8 mGy, 10.5 mGy and 16.3 mGy for the small, medium and large phantoms, respectively. The median AUC obtained from clinical CT protocols was 0.96, 0.90 and 0.83 for the small, medium and large phantoms, respectively.ConclusionsOur study used a model observer to highlight the difference in image quality levels when dealing with the same clinical question. This difference was important and increased with growing phantom size, which generated large variations in patient exposure. In the end, a standardisation initiative may be launched to ensure comparable diagnostic information for well-defined clinical questions. The image quality requirements, related to the clinical question to be answered, should be the starting point of patient dose optimisation.Key Points• Model observers enable to assess image quality objectively based on clinical tasks.• Objective image quality assessment should always include several patient sizes.• Clinical diagnostic image quality should be the starting point for patient dose optimisation.• Dose optimisation by applying DRLs only is insufficient for ensuring clinical requirements.

Characterization of a CT unit for the detection of low contrast structures

Major technological advances in CT enable the acquisition of high quality images while minimizing patient exposure. The goal of this study was to objectively compare two generations of iterative reconstruction (IR) algorithms for the detection of low contrast structures. An abdominal phantom (QRM, Germany), containing 8, 6 and 5mm-diameter spheres (with a nominal contrast of 20HU) was scanned using our standard clinical noise index settings on a GE CT: “Discovery 750 HD”. Two additional rings (2.5 and 5 cm) were also added to the phantom. Images were reconstructed using FBP, ASIR-50%, and VEO (full statistical Model Based Iterative Reconstruction, MBIR). The reconstructed slice thickness was 2.5 mm except 0.625 mm for VEO reconstructions. NPS was calculated to highlight the potential noise reduction of each IR algorithm. To assess LCD (low Contrast Detectability), a Channelized Hotelling Observer (CHO) with 10 DDoG channels was used with the area under the curve (AUC) as a figure of merit. Spheres contrast was also measured. ASIR-50% allowed a noise reduction by a factor two when compared to FBP without an improvement of the LCD. VEO allowed an additional noise reduction with a thinner slice thickness compared to ASIR-50% but with a major improvement of the LCD especially for the large-sized phantom and small lesions. Contrast decreased up to 10% with the phantom size increase for FBP and ASIR-50% and remained constant with VEO. VEO is particularly interesting for LCD when dealing with large patients and small lesion sizes and when the detection task is difficult.