Contrast-detail curves for liver CT

Deep-learning model observer for a low-contrast hepatic metastases localization task in computed tomography

PURPOSE

Conventional model observers (MO) in CT are often limited to a uniform background or varying background that is random and can be modeled in an analytical form. It is unclear if these conventional MOs can be readily generalized to predict human observer performance in clinical CT tasks that involve realistic anatomical background. Deep-learning-based model observers (DL-MO) have recently been developed, but have not been validated for challenging low contrast diagnostic tasks in abdominal CT. We consequently sought to validate a DL-MO for a low-contrast hepatic metastases localization task.

METHODS

We adapted our recently developed DL-MO framework for the liver metastases localization task. Our previously-validated projection-domain lesion-/noise-insertion techniques were used to synthesize realistic positive and low-dose abdominal CT exams, using the archived patient projection data. Ten experimental conditions were generated, which involved different lesion sizes/contrasts, radiation dose levels, and image reconstruction types. Each condition included 100 trials generated from a patient cohort of 7 cases. Each trial was presented as liver image patches (160×160×5 voxels). The DL-MO performance was calculated for each condition and was compared with human observer performance, which was obtained by three sub-specialized radiologists in an observer study. The performance of DL-MO and radiologists was gauged by the area under localization receiver-operating-characteristic curves. The generalization performance of the DL-MO was estimated with the repeated twofold cross-validation method over the same set of trials used in the human observer study. A multi-slice Channelized Hoteling Observers (CHO) was compared with the DL-MO across the same experimental conditions.

RESULTS

The performance of DL-MO was highly correlated to that of radiologists (Pearson's correlation coefficient: 0.987; 95% CI: [0.942, 0.997]). The performance level of DL-MO was comparable to that of the grouped radiologists, that is, the mean performance difference was -3.3%. The CHO performance was poorer than the grouped radiologist performance, before internal noise could be added. The correlation between CHO and radiologists was weaker (Pearson's correlation coefficient: 0.812, and 95% CI: [0.378, 0.955]), and the corresponding performance bias (-29.5%) was statistically significant.

CONCLUSION

The presented study demonstrated the potential of using the DL-MO for image quality assessment in patient abdominal CT tasks.

Deep-learning-based model observer for a lung nodule detection task in computed tomography

Purpose: Task-based image quality assessment using model observers (MOs) is an effective approach to radiation dose and scanning protocol optimization in computed tomography (CT) imaging, once the correlation between MOs and radiologists can be established in well-defined clinically relevant tasks. Conventional MO studies were typically simplified to detection, classification, or localization tasks using tissue-mimicking phantoms, as traditional MOs cannot be readily used in complex anatomical background. However, anatomical variability can affect human diagnostic performance. Approach: To address this challenge, we developed a deep-learning-based MO (DL-MO) for localization tasks and validated in a lung nodule detection task, using previously validated projection-based lesion-/noise-insertion techniques. The DL-MO performance was compared with 4 radiologist readers over 12 experimental conditions, involving varying radiation dose levels, nodule sizes, nodule types, and reconstruction types. Each condition consisted of 100 trials (i.e., 30 images per trial) generated from a patient cohort of 50 cases. DL-MO was trained using small image volume-of-interests extracted across the entire volume of training cases. For each testing trial, the nodule searching of DL-MO was confined to a 3-mm thick volume to improve computational efficiency, and radiologist readers were tasked to review the entire volume. Results: A strong correlation between DL-MO and human readers was observed (Pearson's correlation coefficient: 0.980 with a 95% confidence interval of [0.924, 0.994]). The averaged performance bias between DL-MO and human readers was 0.57%. Conclusion: The experimental results indicated the potential of using the proposed DL-MO for diagnostic image quality assessment in realistic chest CT tasks.

Contrast-to-noise ratio and low-contrast object resolution on full- and low-dose MDCT: SAFIRE versus filtered back projection in a low-contrast object phantom and in the liver

OBJECTIVE

The purpose of this article is to evaluate the effect of sinogram-affirmed iterative reconstruction (SAFIRE) on contrast-to-noise ratio (CNR) compared with filtered back projection (FBP) and to determine whether SAFIRE improves low-contrast object detection or conspicuity in a low-contrast object phantom and in the liver on full- and low-dose examinations.

SUBJECTS AND METHODS

A low-contrast object phantom was scanned at 100%, 70%, 50%, and 30% dose using a single-source made of a dual-source MDCT scanner, with the raw data reconstructed with SAFIRE and FBP. Unenhanced liver CT scans in 22 patients were performed using a dual-source MDCT. The raw data from both tubes (100% dose) were reconstructed using FBP, and data from one tube (50% dose) were reconstructed using both FBP and SAFIRE. CNR was measured in the phantom and in the liver. Noise, contrast, and CNR were compared using paired Student t tests. Six readers assessed sphere detection and conspicuity in the phantom and liver-inferior vena cava conspicuity in the patient data. The phantom and patient data were assessed using multiple-variable logistic regression.

RESULTS

The phantom at 70% and 50% doses with SAFIRE had decreased noise and increased CNR compared with the 100% dose with FBP. In the liver, the mean CNR improvement at 50% dose with SAFIRE compared with FBP was 31.4% and 88% at 100% and 50% doses, respectively (p < 0.001). Sphere object detection and conspicuity improved with SAFIRE (p < 0.001). However, smaller spheres were obscured on both FBP and SAFIRE images at lower doses. Liver-vessel conspicuity improved with SAFIRE over 50%-dose FBP in 67.4% of cases (p < 0.001), and versus 100%-dose FBP, improved in 38.6% of cases (p = 0.085). As a predictor for detection, CNR alone had a discriminatory ability (c-index, 0.970) similar to that of the model that analyzed dose, lesion size, attenuation difference, and reconstruction technique (c-index, 0.978).

CONCLUSION

Lower dose scans reconstructed with SAFIRE have a higher CNR. The ability of SAFIRE to improve low-contrast object detection and conspicuity depends on the radiation dose level. At low radiation doses, low-contrast objects are invisible, regardless of reconstruction technique.

Automatic exposure control at MDCT based on the contrast-to-noise ratio: theoretical background and phantom study

PURPOSE

To develop a new automatic exposure control (AEC) technique based on the contrast-to-noise ratio (CNR) and provide constant lesion detectability.

METHODS

Lesion detectability is affected by factors such as image noise, lesion contrast, and lesion size. We performed ROC analysis to assess the relationship between the optimum CNR and the lesion diameter at various levels of lesion contrast. We then developed a CNR-based AEC algorithm based on lesion detectability. Using CNR- based AEC algorithm, we performed visual evaluation of low-contrast detectability by 5 radiologists on a low-contrast module of the Catphan phantom, a contrast-difference level of 1.0% (difference in the CT number = 10 HU), and objects 3.0-9.0 mm in diameter.

RESULTS

On step-and-shoot scans the mean detection fraction with CNR-based AEC remained almost constant from 88 to 99 % regardless of the lesion size. We observed the same trend on helical scans, the mean detection fraction with CNR-based AEC exhibited a high score from 91 to 100%. Although CNR-based AEC maintains higher CNR for smaller size or lower contrast lesion, radiation dose on 3 mm lesion resulted in about 13 times larger than that of 9 mm lesion size. CTDI(vol) for the CNR-based AEC technique changed dramatically with the SD(Z) from 7.5 to 100.0 mGy for step-and-shoot scans and from 9.1 to 121.5 mGy for helical scans.

CONCLUSIONS

From the viewpoint of ROC analysis-based CNR for lesion detection, CNR-based AEC potentially provide image quality advantages for clinical implementation.

Anatomical background and generalized detectability in tomosynthesis and cone-beam CT

PURPOSE

Anatomical background presents a major impediment to detectability in 2D radiography as well as 3D tomosynthesis and cone-beam CT (CBCT). This article incorporates theoretical and experimental analysis of anatomical background "noise" in cascaded systems analysis of 2D and 3D imaging performance to yield "generalized" metrics of noise-equivalent quanta (NEQ) and detectability index as a function of the orbital extent of the (circular arc) source-detector orbit.

METHODS

A physical phantom was designed based on principles of fractal self-similarity to exhibit power-law spectral density (kappa/Fbeta) comparable to various anatomical sites (e.g., breast and lung). Background power spectra [S(B)(F)] were computed as a function of source-detector orbital extent, including tomosynthesis (approximately 10 degrees -180 degrees) and CBCT (180 degrees + fan to 360 degrees) under two acquisition schemes: (1) Constant angular separation between projections (variable dose) and (2) constant total number of projections (constant dose). The resulting S(B) was incorporated in the generalized NEQ, and detectability index was computed from 3D cascaded systems analysis for a variety of imaging tasks.

RESULTS

The phantom yielded power-law spectra within the expected spatial frequency range, quantifying the dependence of clutter magnitude (kappa) and correlation (beta) with increasing tomosynthesis angle. Incorporation of S(B) in the 3D NEQ provided a useful framework for analyzing the tradeoffs among anatomical, quantum, and electronic noise with dose and orbital extent. Distinct implications are posed for breast and chest tomosynthesis imaging system design-applications varying significantly in kappa and beta, and imaging task and, therefore, in optimal selection of orbital extent, number of projections, and dose. For example, low-frequency tasks (e.g., soft-tissue masses or nodules) tend to benefit from larger orbital extent and more fully 3D tomographic imaging, whereas high-frequency tasks (e.g., microcalcifications) require careful, application-specific selection of orbital extent and number of projections to minimize negative effects of quantum and electronic noise.

CONCLUSIONS

The complex tradeoffs among anatomical background, quantum noise, and electronic noise in projection imaging, tomosynthesis, and CBCT can be described by generalized cascaded systems analysis, providing a useful framework for system design and optimization.

Anniversary paper. Development of x-ray computed tomography: the role of medical physics and AAPM from the 1970s to present

The AAPM, through its members, meetings, and its flagship journal Medical Physics, has played an important role in the development and growth of x-ray tomography in the last 50 years. From a spate of early articles in the 1970s characterizing the first commercial computed tomography (CT) scanners through the "slice wars" of the 1990s and 2000s, the history of CT and related techniques such as tomosynthesis can readily be traced through the pages of Medical Physics and the annals of the AAPM and RSNA/AAPM Annual Meetings. In this article, the authors intend to give a brief review of the role of Medical Physics and the AAPM in CT and tomosynthesis imaging over the last few decades.

Mammographic texture synthesis: second-generation clustered lumpy backgrounds using a genetic algorithm

Synthetic yet realistic images are valuable for many applications in visual sciences and medical imaging. Typically, investigators develop algorithms and adjust their parameters to generate images that are visually similar to real images. In this study, we used a genetic algorithm and an objective, statistical similarity measure to optimize a particular texture generation algorithm, the clustered lumpy backgrounds (CLB) technique, and synthesize images mimicking real mammograms textures. We combined this approach with psychophysical experiments involving the judgment of radiologists, who were asked to qualify the visual realism of the images. Both objective and psychophysical approaches show that the optimized versions are significantly more realistic than the previous CLB model. Anatomical structures are well reproduced, and arbitrary large databases of mammographic texture with visual and statistical realism can be generated. Potential applications include detection experiments, where large amounts of statistically traceable yet realistic images are needed.

Application of the noise power spectrum in modern diagnostic MDCT: part II. Noise power spectra and signal to noise

Balancing dose and image quality requires signal-to-noise (SNR) metrics which incorporate both the variance and the spatial frequency characteristics of noise. In this study, the non-prewhitening matched filter SNR metric is calculated for 2 mm slices of a 1 cm diameter sphere under three different conditions: (1) constant pixel standard deviation, (2) constant dose and (3) constant reconstruction filter. For the constant pixel standard deviation condition, an increase of 260% in SNR was found with increasing filter sharpness. For constant dose, the SNR remained level for smooth to medium filters, then declined by up to 55% with increasing filter sharpness. For a constant reconstruction filter, the SNR increased with dose, but not as high as photon statistics would predict. However, when structured noise was removed from the noise power spectrum, the SNR did vary with quanta statistics. These results offer protocol design guidance for low-frequency-dominated objects.

Gum arabic as a phytochemical construct for the stabilization of gold nanoparticles: in vivo pharmacokinetics and X-ray-contrast-imaging studies

Gold nanoparticles (AuNPs) have exceptional stability against oxidation and therefore will play a significant role in the advancement of clinically useful diagnostic and therapeutic nanomedicines. Despite the huge potential for a new generation of AuNP-based nanomedicinal products, nontoxic AuNP constructs and formulations that can be readily administered site-specifically through the intravenous mode, for diagnostic imaging by computed tomography (CT) or for therapy via various modalities, are still rare. Herein, we report results encompassing: 1) the synthesis and stabilization of AuNPs within the nontoxic phytochemical gum-arabic matrix (GA-AuNPs); 2) detailed in vitro analysis and in vivo pharmacokinetics studies of GA-AuNPs in pigs to gain insight into the organ-specific localization of this new generation of AuNP vector, and 3) X-ray CT contrast measurements of GA-AuNP vectors for potential utility in molecular imaging. Our results demonstrate that naturally occurring GA can be used as a nontoxic phytochemical construct in the production of readily administrable biocompatible AuNPs for diagnostic and therapeutic applications in nanomedicine.

Search for lesions in mammograms: Statistical characterization of observer responses

We investigate human performance for visually detecting simulated microcalcifications and tumors embedded in x-ray mammograms as a function of signal contrast and the number of possible signal locations. Our results show that performance degradation with an increasing number of locations is well approximated by signal detection theory (SDT) with the usual Gaussian assumption. However, more stringent statistical analysis finds a departure from Gaussian assumptions for the detection of microcalcifications. We investigated whether these departures from the SDT Gaussian model could be accounted for by an increase in human internal response correlations arising from the image-pixel correlations present in 1/f spectrum backgrounds and/or observer internal response distributions that departed from the Gaussian assumption. Results were consistent with a departure from the Gaussian response distributions and suggested that the human observer internal responses were more compact than the Gaussian distribution. Finally, we conducted a free search experiment where the signal could appear anywhere within the image. Results show that human performance in a multiple-alternative forced-choice experiment can be used to predict performance in the clinically realistic free search experiment when the investigator takes into account the search area and the observers' inherent spatial imprecision to localize the targets.

Human- and model-observer performance in ramp-spectrum noise: effects of regularization and object variability

We consider detection of a nodule signal profile in noisy images meant to roughly simulate the statistical properties of tomographic image reconstructions in nuclear medicine. The images have two sources of variability arising from quantum noise from the imaging process and anatomical variability in the ensemble of objects being imaged. Both of these sources of variability are simulated by a stationary Gaussian random process. Sample images from this process are generated by filtering white-noise images. Human-observer performance in several signal-known-exactly detection tasks is evaluated through psychophysical studies by using the two-alternative forced-choice method. The tasks considered investigate parameters of the images that influence both the signal profile and pixel-to-pixel correlations in the images. The effect of low-pass filtering is investigated as an approximation to regularization implemented by image-reconstruction algorithms. The relative magnitudes of the quantum and the anatomical variability are investigated as an approximation to the effects of exposure time. Finally, we study the effect of the anatomical correlations in the form of an anatomical slope as an approximation to the effects of different tissue types. Human-observer performance is compared with the performance of a number of model observers computed directly from the ensemble statistics of the images used in the experiments for the purpose of finding predictive models. The model observers investigated include a number of nonprewhitening observers, the Hotelling observer (which is equivalent to the ideal observer for these studies), and six implementations of channelized-Hotelling observers. The human observers demonstrate large effects across the experimental parameters investigated. In the regularization study, performance exhibits a mild peak at intermediate levels of regularization before degrading at higher levels. The exposure-time study shows that human observers are able to detect ever more subtle lesions at increased exposure times. The anatomical slope study shows that human-observer performance degrades as anatomical variability extends into higher spatial frequencies. Of the observers tested, the channelized-Hotelling observers best capture the features of the human data.

Estimation of the noisy component of anatomical backgrounds

The knowledge of the relationship that links radiation dose and image quality is a prerequisite to any optimization of medical diagnostic radiology. Image quality depends, on the one hand, on the physical parameters such as contrast, resolution, and noise, and on the other hand, on characteristics of the observer that assesses the image. While the role of contrast and resolution is precisely defined and recognized, the influence of image noise is not yet fully understood. Its measurement is often based on imaging uniform test objects, even though real images contain anatomical backgrounds whose statistical nature is much different from test objects used to assess system noise. The goal of this study was to demonstrate the importance of variations in background anatomy by quantifying its effect on a series of detection tasks. Several types of mammographic backgrounds and signals were examined by psychophysical experiments in a two-alternative forced-choice detection task. According to hypotheses concerning the strategy used by the human observers, their signal to noise ratio was determined. This variable was also computed for a mathematical model based on the statistical decision theory. By comparing theoretical model and experimental results, the way that anatomical structure is perceived has been analyzed. Experiments showed that the observer's behavior was highly dependent upon both system noise and the anatomical background. The anatomy partly acts as a signal recognizable as such and partly as a pure noise that disturbs the detection process. This dual nature of the anatomy is quantified. It is shown that its effect varies according to its amplitude and the profile of the object being detected. The importance of the noisy part of the anatomy is, in some situations, much greater than the system noise. Hence, reducing the system noise by increasing the dose will not improve task performance. This observation indicates that the tradeoff between dose and image quality might be optimized by accepting a higher system noise. This could lead to a better resolution, more contrast, or less dose.

Detecting hepatic lesions: the added utility of CT liver window settings

PURPOSE

To prospectively evaluate the utility of adding computed tomographic (CT) liver windows to conventional soft-tissue windows for the detection of hepatic disease.

MATERIALS AND METHODS

One of four radiologists experienced in abdominal imaging interpreted 1,175 consecutive abdominal CT scans from one institution. Hepatic images were first interpreted by using standard soft-tissue windows. The number of lesions and confidence in lesion detection were recorded. The liver-window images were then interpreted in conjunction with the soft-tissue-window images, and the number of lesions and confidence in detection were recorded again. The proportion of patients in whom additional lesions were found by using liver windows was determined.

RESULTS

On soft-tissue-window and liver-window scans interpreted together, 869 (74%) patients had no hepatic lesions. Thirty-six (3.1%) patients had new lesions seen with the addition of liver windows. Twelve of these 36 patients had no lesions seen on soft-tissue-window scans. Twenty-six of the 36 patients with additional lesions seen had a history of neoplasm. There was a change in diagnosis in 1.7% of the patients with the addition of liver windows and a change in recommendation for follow-up in 0.85%.

CONCLUSION

Routine interpretation of liver-window scans for all abdominal CT scans has limited added utility in detecting hepatic disease.

Influence of visual distractors on detectability of liver nodules on contrast-enhanced spiral computed tomography scans

RATIONALE AND OBJECTIVES

The authors evaluated the ability of observers to identify simulated nodules placed electronically on normal contrast material-enhanced computed tomography (CT) scans of the liver to assess the effect of nodule size and polarity on detection and localization.

METHODS

Seven readers evaluated two sets of CT scans that contained 80 stimuli each. The simulated nodules were either darker or brighter than the contrast-enhanced liver and were 5.6-8.0 mm in diameter. Readers were asked to find the most suspicious-looking nodule on each section and rate the likelihood that the chosen location actually contained a nodule.

RESULTS

The fraction of nodules found by each observer was substantially greater for dark nodules than for bright ones (0.679 +/- 0.03 vs 0.345 +/- 0.045, respectively [mean +/- standard error]). This difference was consistent for all nodule sizes. Additional analyses (including receiver operating characteristic curves of conditional responses) suggested that the presence of bright blood vessels distracted the readers and decreased their ability to find bright nodules.

CONCLUSION

Normal vascular structures on contrast-enhanced CT scans of the liver impair an observer's ability to detect bright liver nodules.

Comparison of trained and untrained observers using subjective and objective measures of imaging performance

RATIONALE AND OBJECTIVES

We compared subjective and objective measures of imaging performance using variations of Rose- and Burger-type low-contrast phantoms with trained (radiology residents) and untrained (graduate students) observers.

METHODS

With one phantom variant, observers indicated the total number of objects seen when test objects were presented in a regular pattern (subjective). With the second phantom variant, observers stated whether a low-contrast disk was present in each locale, thereby permitting the true-positive fraction and false-positive fraction to be determined (objective).

RESULTS

The untrained-observer group had a significantly lower imaging performance than the trained observer group in subjective tests. These differences were not found on objective tests. For the trained-observer group, similar contrast levels were required in subjective and objective tests to yield a 50% rate of detection.

CONCLUSION

Trained observers are superior subjects compared with untrained observers for assessing imaging performance using subjective low-contrast phantoms. In experiments using phantoms that allowed objective testing, both groups of observers yield similar results.