Daniel J. Blezek

Adaptive nonlocal means filtering based on local noise level for CT denoising

PURPOSE To develop and evaluate an image-domain noise reduction method based on a modified nonlocal means (NLM) algorithm that is adaptive to local noise level of CT images and to implement this method in a time frame consistent with clinical workflow. METHODS A computationally efficient technique for local noise estimation directly from CT images was developed. A forward projection, based on a 2D fan-beam approximation, was used to generate the projection data, with a noise model incorporating the effects of the bowtie filter and automatic exposure control. The noise propagation from projection data to images was analytically derived. The analytical noise map was validated using repeated scans of a phantom. A 3D NLM denoising algorithm was modified to adapt its denoising strength locally based on this noise map. The performance of this adaptive NLM filter was evaluated in phantom studies in terms of in-plane and cross-plane high-contrast spatial resolution, noise power spectrum (NPS), subjective low-contrast spatial resolution using the American College of Radiology (ACR) accreditation phantom, and objective low-contrast spatial resolution using a channelized Hotelling model observer (CHO). Graphical processing units (GPU) implementation of this noise map calculation and the adaptive NLM filtering were developed to meet demands of clinical workflow. Adaptive NLM was piloted on lower dose scans in clinical practice. RESULTS The local noise level estimation matches the noise distribution determined from multiple repetitive scans of a phantom, demonstrated by small variations in the ratio map between the analytical noise map and the one calculated from repeated scans. The phantom studies demonstrated that the adaptive NLM filter can reduce noise substantially without degrading the high-contrast spatial resolution, as illustrated by modulation transfer function and slice sensitivity profile results. The NPS results show that adaptive NLM denoising preserves the shape and peak frequency of the noise power spectrum better than commercial smoothing kernels, and indicate that the spatial resolution at low contrast levels is not significantly degraded. Both the subjective evaluation using the ACR phantom and the objective evaluation on a low-contrast detection task using a CHO model observer demonstrate an improvement on low-contrast performance. The GPU implementation can process and transfer 300 slice images within 5 min. On patient data, the adaptive NLM algorithm provides more effective denoising of CT data throughout a volume than standard NLM, and may allow significant lowering of radiation dose. After a two week pilot study of lower dose CT urography and CT enterography exams, both GI and GU radiology groups elected to proceed with permanent implementation of adaptive NLM in their GI and GU CT practices. CONCLUSIONS This work describes and validates a computationally efficient technique for noise map estimation directly from CT images, and an adaptive NLM filtering based on this noise map, on phantom and patient data. Both the noise map calculation and the adaptive NLM filtering can be performed in times that allow integration with clinical workflow. The adaptive NLM algorithm provides effective denoising of CT data throughout a volume, and may allow significant lowering of radiation dose.

Methods for Clinical Evaluation of Noise Reduction Techniques in Abdominopelvic CT

Most noise reduction methods involve nonlinear processes, and objective evaluation of image quality can be challenging, since image noise cannot be fully characterized on the sole basis of the noise level at computed tomography (CT). Noise spatial correlation (or noise texture) is closely related to the detection and characterization of low-contrast objects and may be quantified by analyzing the noise power spectrum. High-contrast spatial resolution can be measured using the modulation transfer function and section sensitivity profile and is generally unaffected by noise reduction. Detectability of low-contrast lesions can be evaluated subjectively at varying dose levels using phantoms containing low-contrast objects. Clinical applications with inherent high-contrast abnormalities (eg, CT for renal calculi, CT enterography) permit larger dose reductions with denoising techniques. In low-contrast tasks such as detection of metastases in solid organs, dose reduction is substantially more limited by loss of lesion conspicuity due to loss of low-contrast spatial resolution and coarsening of noise texture. Existing noise reduction strategies for dose reduction have a substantial impact on lowering the radiation dose at CT. To preserve the diagnostic benefit of CT examination, thoughtful utilization of these strategies must be based on the inherent lesion-to-background contrast and the anatomy of interest. The authors provide an overview of existing noise reduction strategies for low-dose abdominopelvic CT, including analytic reconstruction, image and projection space denoising, and iterative reconstruction; review qualitative and quantitative tools for evaluating these strategies; and discuss the strengths and limitations of individual noise reduction methods.

Optimizing non-local means for denoising low dose CT

Due to the rapid increase in use of CT imaging and the recently-heightened awareness of radiation-induced cancer, improving the diagnostic quality of low dose CT has become increasingly important. One potential method is to increase the signal-to-noise ratio of low dose images through denoising. Non-local means is a promising approach; however, it has many potentially adjustable parameters and application-specific areas of improvement. The filter uses a weighted average of similar regions to denoise each image pixel. Though the classic formulation uses only patches from the image being filtered, these patches can, in principle, be drawn from other images. In CT images, patches can be drawn from neighboring slices. We used that potential to increase the peak signal-to-noise ratio (PSNR) by over 4 dB when denoising low dose phantom CT images, and quantitatively demonstrated the filters sensitivity to adjustment of each of its parameters.

Estimating amounts of iron oxide from gradient echo images

Rat legs directly injected with superparamagnetic iron oxide (SPIO) were studied by dual‐echo, gradient‐echo imaging. The amount of iron injected was estimated using a point dipole model for the SPIO injection site. Saturation magnetization of 6:1 PEG/amino modified silane‐coated iron oxide particles with 5‐ to 6‐nm core and 20–25 hydrodynamic diameter was ∼110 emu/g of iron. Estimates of the amount of iron injected made from signal void volumes surrounding SPIO centers yielded erroneous results varying with sample orientation in the scanner and echo time (TE). For example, a 10 μL, 3‐μg iron injection produced signal void volumes of 80 and 210 μL at TE of 9.8 and 25 ms, respectively, giving apparent iron contents of 6 ± 1 and 10 ± 2 μg respectively. A more effective approach uses the phase difference between two gradient recalled echo images. To estimate iron content, this approach fits the expected (3 cos2θ − 1)/|r|3 spatial phase distribution to the observed phase differences. Extraneous phase effects made fitting phase at a single TE ineffective. With the dual echo method, 18 independent estimates were 2.48 ± 0.26 μg std, independently of sample orientation. Estimates in empty control regions were −90 and −140 ng. A 1‐μg injection indicated 0.5, 1.2, and 1.2 μg. Magn Reson Med, 2009.

Observer Performance in the Detection and Classification of Malignant Hepatic Nodules and Masses with CT Image-Space Denoising and Iterative Reconstruction

PURPOSE To determine if lower-dose computed tomographic (CT) scans obtained with adaptive image-based noise reduction (adaptive nonlocal means [ANLM]) or iterative reconstruction (sinogram-affirmed iterative reconstruction [SAFIRE]) result in reduced observer performance in the detection of malignant hepatic nodules and masses compared with routine-dose scans obtained with filtered back projection (FBP). MATERIALS AND METHODS This study was approved by the institutional review board and was compliant with HIPAA. Informed consent was obtained from patients for the retrospective use of medical records for research purposes. CT projection data from 33 abdominal and 27 liver or pancreas CT examinations were collected (median volume CT dose index, 13.8 and 24.0 mGy, respectively). Hepatic malignancy was defined by progression or regression or with histopathologic findings. Lower-dose data were created by using a validated noise insertion method (10.4 mGy for abdominal CT and 14.6 mGy for liver or pancreas CT) and images reconstructed with FBP, ANLM, and SAFIRE. Four readers evaluated routine-dose FBP images and all lower-dose images, circumscribing liver lesions and selecting diagnosis. The jackknife free-response receiver operating characteristic figure of merit (FOM) was calculated on a per-malignant nodule or per-mass basis. Noninferiority was defined by the lower limit of the 95% confidence interval (CI) of the difference between lower-dose and routine-dose FOMs being less than -0.10. RESULTS Twenty-nine patients had 62 malignant hepatic nodules and masses. Estimated FOM differences between lower-dose FBP and lower-dose ANLM versus routine-dose FBP were noninferior (difference: -0.041 [95% CI: -0.090, 0.009] and -0.003 [95% CI: -0.052, 0.047], respectively). In patients with dedicated liver scans, lower-dose ANLM images were noninferior (difference: +0.015 [95% CI: -0.077, 0.106]), whereas lower-dose FBP images were not (difference -0.049 [95% CI: -0.140, 0.043]). In 37 patients with SAFIRE reconstructions, the three lower-dose alternatives were found to be noninferior to the routine-dose FBP. CONCLUSION At moderate levels of dose reduction, lower-dose FBP images without ANLM or SAFIRE were noninferior to routine-dose images for abdominal CT but not for liver or pancreas CT.

Anesthesiology training using 3D imaging and virtual reality

Current training for regional nerve block procedures by anesthesiology residents requires expert supervision and the use of cadavers; both of which are relatively expensive commodities in todays cost-conscious medical environment. We are developing methods to augment and eventually replace these training procedures with real-time and realistic computer visualizations and manipulations of the anatomical structures involved in anesthesiology procedures, such as nerve plexus injections (e.g., celiac blocks). The initial work is focused on visualizations: both static images and rotational renderings. From the initial results, a coherent paradigm for virtual patient and scene representation will be developed.

Towards a More Cloud-Friendly Medical Imaging Applications Architecture: A Modest Proposal

Recent information technology literature, in general, and radiology trade journals, in particular, are rife with allusions to the “cloud” suggesting that moving one’s compute and storage assets into someone else’s data center magically solves cost, performance, and elasticity problems. More likely, one is only trading one set of problems for another, including greater latency (aka slower turnaround times) since the image data must now leave the local area network and travel longer paths via encrypted tunnels. To offset this, an imaging system design is needed that reduces the number of high-latency image transmissions, yet can still leverage cloud strengths. This work explores the requirements for such a design.

Haptic rendering of isosurfaces directly from medical images.

Virtual environments for surgical training, planning and rehearsal have the potential to significantly enhance patient treatment and diagnosis. Haptic feedback devices provide forces to the physician through a manipulator, simulating palpation, scalpel cuts, or retraction of tissue. While haptics have been studied in other fields, medical applications of haptics remain in their infancy. We propose a method of haptically rendering isosurfaces (representing hard structures) directly from anatomical datasets rather than through traditional intermediate graphical representations such as polygons. Our algorithm determines an implicit surface representation of a volumetric isosurface on the fly, and renders the structure using standard haptic algorithms to calculate the forces felt by the user. This approach has the advantage of providing easy access to the rich volume dataset containing the actual anatomy. By relating Hounsfield units to density, haptic rendering has the ability to provide different resistances based on the tissue type being rendered. We developed and tested our algorithm using a quadric implicit surface, a common primitive in computer graphics, and have applied it to a variety of anatomical image datasets. Our paper describes the algorithm in sufficient detail to facilitate reproduction by others.

Application of the 3D slicer chest imaging platform segmentation algorithm for large lung nodule delineation

Purpose Accurate segmentation of lung nodules is crucial in the development of imaging biomarkers for predicting malignancy of the nodules. Manual segmentation is time consuming and affected by inter-observer variability. We evaluated the robustness and accuracy of a publically available semiautomatic segmentation algorithm that is implemented in the 3D Slicer Chest Imaging Platform (CIP) and compared it with the performance of manual segmentation. Methods CT images of 354 manually segmented nodules were downloaded from the LIDC database. Four radiologists performed the manual segmentation and assessed various nodule characteristics. The semiautomatic CIP segmentation was initialized using the centroid of the manual segmentations, thereby generating four contours for each nodule. The robustness of both segmentation methods was assessed using the region of uncertainty (δ) and Dice similarity index (DSI). The robustness of the segmentation methods was compared using the Wilcoxon-signed rank test (pWilcoxon<0.05). The Dice similarity index (DSIAgree) between the manual and CIP segmentations was computed to estimate the accuracy of the semiautomatic contours. Results The median computational time of the CIP segmentation was 10 s. The median CIP and manually segmented volumes were 477 ml and 309 ml, respectively. CIP segmentations were significantly more robust than manual segmentations (median δCIP = 14ml, median dsiCIP = 99% vs. median δmanual = 222ml, median dsimanual = 82%) with pWilcoxon~10−16. The agreement between CIP and manual segmentations had a median DSIAgree of 60%. While 13% (47/354) of the nodules did not require any manual adjustment, minor to substantial manual adjustments were needed for 87% (305/354) of the nodules. CIP segmentations were observed to perform poorly (median DSIAgree≈50%) for non-/sub-solid nodules with subtle appearances and poorly defined boundaries. Conclusion Semi-automatic CIP segmentation can potentially reduce the physician workload for 13% of nodules owing to its computational efficiency and superior stability compared to manual segmentation. Although manual adjustment is needed for many cases, CIP segmentation provides a preliminary contour for physicians as a starting point.

MIRMAID: A Content Management System for Medical Image Analysis Research.

Today, a typical clinical study can involve thousands of participants, with imaging data acquired over several time points across multiple institutions. The additional associated information (metadata) accompanying these data can cause data management to be a study-hindering bottleneck. Consistent data management is crucial for large-scale modern clinical imaging research studies. If the study is to be used for regulatory submissions, such systems must be able to meet regulatory compliance requirements for systems that manage clinical image trials, including protecting patient privacy. Our aim was to develop a system to address these needs by leveraging the capabilities of an open-source content management system (CMS) that has a highly configurable workflow; has a single interface that can store, manage, and retrieve imaging-based studies; and can handle the requirement for data auditing and project management. We developed a Web-accessible CMS for medical images called Medical Imaging Research Management and Associated Information Database (MIRMAID). From its inception, MIRMAID was developed to be highly flexible and to meet the needs of diverse studies. It fulfills the need for a complete system for medical imaging research management.