Standardization in Quantitative Imaging: A Multicenter Comparison of Radiomic Features from Different Software Packages on Digital Reference Objects and Patient Data Sets.

Radiomic features are being increasingly studied for clinical applications. We aimed to assess the agreement among radiomic features when computed by several groups by using different software packages under very tightly controlled conditions, which included standardized feature definitions and common image data sets. Ten sites (9 from the NCI's Quantitative Imaging Network] positron emission tomography-computed tomography working group plus one site from outside that group) participated in this project. Nine common quantitative imaging features were selected for comparison including features that describe morphology, intensity, shape, and texture. The common image data sets were: three 3D digital reference objects (DROs) and 10 patient image scans from the Lung Image Database Consortium data set using a specific lesion in each scan. Each object (DRO or lesion) was accompanied by an already-defined volume of interest, from which the features were calculated. Feature values for each object (DRO or lesion) were reported. The coefficient of variation (CV), expressed as a percentage, was calculated across software packages for each feature on each object. Thirteen sets of results were obtained for the DROs and patient data sets. Five of the 9 features showed excellent agreement with CV < 1%; 1 feature had moderate agreement (CV < 10%), and 3 features had larger variations (CV ≥ 10%) even after attempts at harmonization of feature calculations. This work highlights the value of feature definition standardization as well as the need to further clarify definitions for some features.

Radiomics in Brain Tumor: Image Assessment, Quantitative Feature Descriptors, and Machine-Learning Approaches.

Radiomics describes a broad set of computational methods that extract quantitative features from radiographic images. The resulting features can be used to inform imaging diagnosis, prognosis, and therapy response in oncology. However, major challenges remain for methodologic developments to optimize feature extraction and provide rapid information flow in clinical settings. Equally important, to be clinically useful, predictive radiomic properties must be clearly linked to meaningful biologic characteristics and qualitative imaging properties familiar to radiologists. Here we use a cross-disciplinary approach to highlight studies in radiomics. We review brain tumor radiologic studies (eg, imaging interpretation) through computational models (eg, computer vision and machine learning) that provide novel clinical insights. We outline current quantitative image feature extraction and prediction strategies with different levels of available clinical classes for supporting clinical decision-making. We further discuss machine-learning challenges and data opportunities to advance radiomic studies.

Measurement of cardiac output by computed transmission tomography.

The capability of computed tomography (CT) scanning to measure cardiac output was explored using ten anesthetized dogs, and the results were compared with those obtained by thermodilution. Dynamic CT scans were performed at the level of the aortic root while small peripheral intravenous boluses of contrast medium were injected. Time/density curves were generated using a gamma variate fitting program. These were used to estimate cardiac output by applying indicator dilution principles. CT results correlated favorably (r = 0.86) with those of thermodilution. This feasibility study indicates the utility of CT for obtaining physiologic measurements of cardiac function and should encourage further studies to develop the potential of CT for cardiovascular diagnostic purposes.

Single breath-hold pulmonary magnetic resonance angiography. Optimization and comparison of three imaging strategies.

RATIONALE AND OBJECTIVES: Ultrafast gradient-recalled-echo techniques for obtaining high-quality pulmonary magnetic resonance angiograms within a single breath-hold were optimized. METHODS: Fourteen subjects were imaged with both the body coil and a phased-array surface coil, using three gradient-recalled-echo pulse sequences: 1) two-dimensional sequential; 2) two-dimensional interleaved; and 3) volumetric acquisitions. Image quality was assessed with varied flip angle, receiver bandwidth, slice thickness/number, and matrix size. Cardiac compensation diminished ghost artifacts in the interleaved sequence. Individual sagittal sections and maximum intensity projections were reviewed. RESULTS: Pulmonary magnetic resonance angiograms acquired with volumetric and two-dimensional interleaved gradient-recalled-echo pulse sequences benefit greatest from intravenous gadolinium and result in greater pulmonary arterial visualization than traditional time-of-flight techniques. Phased-array coils result in improved vessel detection. CONCLUSIONS: High-quality breath-held pulmonary magnetic resonance angiography can be obtained with an intravenous contrast-enhanced gradient-recalled-echo acquisition; however, image quality is dependent on the pulse sequence.

Spatially varying longitudinal aliasing and resolution in spiral computed tomography.

Spiral computed tomography (CT) has revolutionized conventional CT as a truly three-dimensional imaging modality. A number of studies aimed at evaluating the longitudinal resolution in spiral CT have been presented, but the spatially varying nature of the longitudinal resolution in spiral CT has been largely left undiscussed. In this paper, we investigate the longitudinal resolution in spiral CT as affected by the spatially varying longitudinal aliasing. We propose the treatment of aliasing as a signal dependent, additive noise, and define a new image quality parameter, the contrast-to-aliased-noise ratio (CNaR), that relates to possible image degradation or loss of resolution caused by aliasing. We performed CT simulations and actual phantom scans using a resolution phantom consisting of sequences of spherical beads of different diameters, extending along the longitudinal axis. Our results show that the off-isocenter longitudinal resolution differs significantly from the longitudinal resolution at the isocenter and that the CNaR decreases with distance from the isocenter, and is a function of pitch and the helical interpolation algorithm used. The longitudinal resolution was observed to worsen with decreasing CNaR. We conclude that the longitudinal resolution in spiral CT is spatially varying, and can be characterized by the CNaR measured at the transaxial location of interest.

Automated flight path planning for virtual endoscopy.

In this paper, a novel technique for rapid and automatic computation of flight paths for guiding virtual endoscopic exploration of three-dimensional medical images is described. While manually planning flight paths is a tedious and time consuming task, our algorithm is automated and fast. Our method for positioning the virtual camera is based on the medial axis transform but is much more computationally efficient. By iteratively correcting a path toward the medial axis, the necessity of evaluating simple point criteria during morphological thinning is eliminated. The virtual camera is also oriented in a stable viewing direction, avoiding sudden twists and turns. We tested our algorithm on volumetric data sets of eight colons, one aorta and one bronchial tree. The algorithm computed the flight paths in several minutes per volume on an inexpensive workstation with minimal computation time added for multiple paths through branching structures (10%-13% per extra path). The results of our algorithm are smooth, centralized paths that aid in the task of navigation in virtual endoscopic exploration of three-dimensional medical images.

Modeling of polychromatic attenuation using computed tomography reconstructed images.

This paper presents a procedure for estimating an accurate model of the CT imaging process including spectral effects. As raw projection data are typically unavailable to the end-user, we adopt a post-processing approach that utilizes the reconstructed images themselves. This approach includes errors from x-ray scatter and the nonidealities of the built-in soft tissue correction into the beam characteristics, which is crucial to beam hardening correction algorithms that are designed to be applied directly to CT reconstructed images. We formulate this approach as a quadratic programming problem and propose two different methods, dimension reduction and regularization, to overcome ill conditioning in the model. For the regularization method we use a statistical procedure, Cross Validation, to select the regularization parameter. We have constructed step-wedge phantoms to estimate the effective beam spectrum of a GE CT-I scanner. Using the derived spectrum, we computed the attenuation ratios for the wedge phantoms and found that the worst case modeling error is less than 3% of the corresponding attenuation ratio. We have also built two test (hybrid) phantoms to evaluate the effective spectrum. Based on these test phantoms, we have shown that the effective beam spectrum provides an accurate model for the CT imaging process. Last, we used a simple beam hardening correction experiment to demonstrate the effectiveness of the estimated beam profile for removing beam hardening artifacts. We hope that this estimation procedure will encourage more independent research on beam hardening corrections and will lead to the development of application-specific beam hardening correction algorithms.

Registration error quantification of a surface-based multimodality image fusion system.

This paper presents a new reference data set and associated quantification methodology to assess the accuracy of registration of computerized tomography (CT) and magnetic-resonance (MR) images. Also described is a new semiautomatic surface-based system for registering and visualizing CT and MR images. The registration error of the system was determined using a reference data set that was obtained from a cadaver in which rigid fiducial tubes were inserted prior to imaging. Registration error was measured as the distance between an analytic expression for each fiducial tube in one image set and transformed samples of the corresponding tube obtained from the other. Registration was accomplished by first identifying surfaces of similar anatomic structures in each image set. A transformation that best registered these structures was determined using a nonlinear optimization procedure. Even though the root-mean-square (rms) distance at the registered surfaces was similar to that reported by other groups, it was found that rms distances for the tubes were significantly larger than the final rms distances between the registered surfaces. It was also found that minimizing rms distance at the surface did not minimize rms distance for the tubes.

A new frame-based registration algorithm.

This paper presents a new algorithm for frame registration. Our algorithm requires only that the frame be comprised of straight rods, as opposed to the N structures or an accurate frame model required by existing algorithms. The algorithm utilizes the full 3D information in the frame as well as a least squares weighting scheme to achieve highly accurate registration. We use simulated CT data to assess the accuracy of our algorithm. We compare the performance of the proposed algorithm to two commonly used algorithms. Simulation results show that the proposed algorithm is comparable to the best existing techniques with knowledge of the exact mathematical frame model. For CT data corrupted with an unknown in-plane rotation or translation, the proposed technique is also comparable to the best existing techniques. However, in situations where there is a discrepancy of more than 2 mm (0.7% of the frame dimension) between the frame and the mathematical model, the proposed technique is significantly better (p < or = 0.05) than the existing techniques. The proposed algorithm can be applied to any existing frame without modification. It provides better registration accuracy and is robust against model mis-match. It allows greater flexibility on the frame structure. Lastly, it reduces the frame construction cost as adherence to a concise model is not required.

Characterization of spatial distortion in magnetic resonance imaging and its implications for stereotactic surgery.

The different sources of spatial distortion in magnetic resonance images are reviewed from the point of view of stereotactic target localization. The extents of the two most complex sources of spatial distortion, gradient field nonlinearities and magnetic field inhomogeneities, are discussed both qualitatively and quantitatively. Several ways by which the spatial distortion resulting from these sources can be minimized are discussed. The clinical relevance of the spatial distortion along with some strategies to minimize the localization errors in magnetic resonance-guided stereotaxy are presented.

Medical image segmentation using analysis of isolable-contour maps.

A common challenge for automated segmentation techniques is differentiation between images of close objects that have similar intensities, whose boundaries are often blurred due to partial-volume effects. We propose a novel approach to segmentation of two-dimensional images, which addresses this challenge. Our method, which we call intrinsic shape for segmentation (ISeg), analyzes isolabel-contour maps to identify coherent regions that correspond to major objects. ISeg generates an isolabel-contour map for an image by multilevel thresholding with a fine partition of the intensity range. ISeg detects object boundaries by comparing the shape of neighboring isolabel contours from the map. ISeg requires only little effort from users; it does not require construction of shape models of target objects. In a formal validation with computed-tomography angiography data, we showed that ISeg was more robust than conventional thresholding, and that ISeg's results were comparable to results of manual tracing.

Longitudinal sampling and aliasing in spiral CT.

Although analyses of in-plane aliasing have been done for conventional computed tomography (CT) images, longitudinal aliasing in spiral CT has not been properly investigated. We propose a mathematical model of the three-dimensional (3-D) sampling scheme in spiral CT and analyze its effects on longitudinal aliasing. We investigated longitudinal aliasing as a function of the helical-interpolation algorithm, pitch, and reconstruction interval using CT simulations and actual phantom scans. Our model predicts, and we verified, that for a radially uniform object at the isocenter, the spiral sampling scheme results in spatially varying cancellation of the aliased spectral islands which, in turn, results in spatially varying longitudinal aliasing. The aliasing is minimal at the scanner isocenter, but worsens with distance from it and rapidly becomes significant. Our results agree with published results observed at the isocenter of the scanner and further provide new insight into the aliasing conditions at off-isocenter locations with respect to the pitch, interpolation algorithm, and reconstruction interval. We conclude that longitudinal aliasing at off-isocenter locations can be significant, and that its magnitude and effects cannot be predicted by measurements made only at the scanner isocenter.

Reconstruction algorithm for polychromatic CT imaging: application to beam hardening correction.

This paper presents a new reconstruction algorithm for both single- and dual-energy computed tomography (CT) imaging. By incorporating the polychromatic characteristics of the X-ray beam into the reconstruction process, the algorithm is capable of eliminating beam hardening artifacts. The single energy version of the algorithm assumes that each voxel in the scan field can be expressed as a mixture of two known substances, for example, a mixture of trabecular bone and marrow, or a mixture of fat and flesh. These assumptions are easily satisfied in a quantitative computed tomography (QCT) setting. We have compared our algorithm to three commonly used single-energy correction techniques. Experimental results show that our algorithm is much more robust and accurate. We have also shown that QCT measurements obtained using our algorithm are five times more accurate than that from current QCT systems (using calibration). The dual-energy mode does not require any prior knowledge of the object in the scan field, and can be used to estimate the attenuation coefficient function of unknown materials. We have tested the dual-energy setup to obtain an accurate estimate for the attenuation coefficient function of K2 HPO4 solution.

Determining cardiac velocity fields and intraventricular pressure distribution from a sequence of ultrafast CT cardiac images.

A method of computing the velocity field and pressure distribution from a sequence of ultrafast CT (UFCT) cardiac images is demonstrated. UFCT multi-slice cine imaging gives a series of tomographic slices covering the volume of the heart at a rate of 17 frames per second. The complete volume data set can be modeled using equations of continuum theory and through regularization, velocity vectors of both blood and tissue can be determined at each voxel in the volume. The authors present a technique to determine the pressure distribution throughout the volume of the left ventricle using the computed velocity field. A numerical algorithm is developed by discretizing the pressure Poisson equation (PPE), which Is based on the Navier-Stokes equation. The algorithm is evaluated using a mathematical phantom of known velocity and pressure-Couette flow. It is shown that the algorithm based on the PPE can reconstruct the pressure distribution using only the velocity data. Furthermore, the PPE is shown to be robust in the presence of noise. The velocity field and pressure distribution derived from a UFCT study of a patient are also presented.

A statistical 3-D pattern processing method for computer-aided detection of polyps in CT colonography.

Adenomatous polyps in the colon are believed to be the precursor to colorectal carcinoma, the second leading cause of cancer deaths in United States. In this paper, we propose a new method for computer-aided detection of polyps in computed tomography (CT) colonography (virtual colonoscopy), a technique in which polyps are imaged along the wall of the air-inflated, cleansed colon with X-ray CT. Initial work with computer aided detection has shown high sensitivity, but at a cost of too many false positives. We present a statistical approach that uses support vector machines to distinguish the differentiating characteristics of polyps and healthy tissue, and uses this information for the classification of the new cases. One of the main contributions of the paper is the new three-dimensional pattern processing approach, called random orthogonal shape sections method, which combines the information from many random images to generate reliable signatures of shape. The input to the proposed system is a collection of volume data from candidate polyps obtained by a high-sensitivity, low-specificity system that we developed previously. The results of our ten-fold cross-validation experiments show that, on the average, the system increases the specificity from 0.19 (0.35) to 0.69 (0.74) at a sensitivity level of 1.0 (0.95).

MRI of pulmonary embolism using Gd-DTPA-polyethylene glycol polymer enhanced 3D fast gradient echo technique in a canine model.

This study was to evaluate the accuracy of MR angiography (MRA) using a Gd-DTPA-polyethylene glycol polymer (Gd-DTPA-PEG) with a 3D fast gradient echo (3D fgre) technique in diagnosing pulmonary embolism in a canine model. Pulmonary emboli were created in six mongrel dogs (20-30 kg) by injecting tantalum oxide-doped autologous blood clots into the femoral veins via cutdowns. MRI was performed with a 1.5 T GE Signa imager using a 3D fgre sequence (11.9/2.3/15 degrees) following intravenous injection of 0.06 mmol Gd/kg of Gd-DTPA-PEG. The dogs were euthanized and spiral CT of the lungs were then obtained on the deceased dogs. The MRI images were reviewed independently and receiver-operating-characteristic (ROC) curves were used for statistical analysis using spiral CT results as the gold standard. The pulmonary emboli were well visualized on spiral CT. Out of 108 pulmonary segments in the six dogs, 24 contained emboli >2 mm and 27 contained emboli < or = 2 mm. With unblinded review, MRI detected 79% of emboli >2 mm and only 48% of emboli < or = 2 mm. The blinded review results were significantly worse. Gd-DTPA-PEG enhanced 3D fgre MRI is potentially able to demonstrate pulmonary embolism with fairly high degree of accuracy, but specialized training for the interpretations will be required.

Virtual endoscopy of the paranasal sinuses using perspective volume rendered helical sinus computed tomography.

Our goal was to use three-dimensional information obtained from helical computed tomographic (CT) data to explore and evaluate the nasal cavity, nasopharynx, and paranasal sinuses by simulated virtual endoscopy (VE). This was done by utilizing a new image reconstruction method known as perspective volume rendering (PVR). Thin-section helical CT of the nasal cavity, nasopharynx, and paranasal sinuses was performed on a conventional CT scanner. The data were transferred to a workstation to create views similar to those seen with endoscopy. Additional views not normally accessible by conventional endoscopy were generated. Key perspectives were selected, and a video "flight" model was choreographed and synthesized through the nasal cavity and sinuses based on the CT data. VE allows evaluation of the nasal cavity, nasopharynx, and paranasal sinuses with appreciation of the relationships of these spatially complex structures. In addition, this technique allows structural visualization with unconventional angles, perspectives, and locations not conventionally accessible. Although biopsies, cultures, and lavages routinely done with endoscopy cannot be performed with VE, this technique holds promise for improving the diagnostic evaluation of the nasal cavity, the nasopharynx, and the paranasal sinuses. The unconventional visual perspectives and very low morbidity may complement many applications of simple diagnostic endoscopy.

Phase unwrapping of MR phase images using Poisson equation.

The authors have developed a technique based on a solution of the Poisson equation to unwrap the phase in magnetic resonance (MR) phase images. The method is based on the assumption that the magnitude of the inter-pixel phase change is less than pi per pixel. Therefore, the authors obtain an estimate of the phase gradient by "wrapping" the gradient of the original phase image. The problem is then to obtain the absolute phase given the estimate of the phase gradient. The least-squares (LS) solution to this problem is shown to be a solution of the Poisson equation allowing the use of fast Poisson solvers. The absolute phase is then obtained by mapping the LS phase to the nearest multiple of 2 K from the measured phase. The proposed technique is evaluated using MR phase images and is proven to be robust in the presence of noise. An application of the proposed method to the 3-point Dixon technique for water and fat separation is demonstrated.

Interactive display of volumetric data by Fast Fourier Projection.

This article describes a new algorithm for reprojection of volumetric data, called Fast Fourier Projection (FFP), which is one to two orders of magnitude faster than conventional methods such as ray casting. The theoretical basis of the new method is developed in a unified mathematical framework encompassing slice imaging and conventional volumetric reprojection methods. Software implementation is discussed in detail. The article closes with an account of experience with a prototype FFP implementation, and applications of the technique in medical visualization.

Imaging informatics: toward capturing and processing semantic information in radiology images.

OBJECTIVES: To identify challenges and opportunities in imaging informatics that can lead to the use of images for discovery, and that can potentially improve the diagnostic accuracy of imaging professionals. METHODS: Recent articles on imaging informatics and related articles from PubMed were reviewed and analyzed. Some new developments and challenges that recent research in imaging informatics will meet are identified and discussed. RESULTS: While much literature continues to be devoted to traditional imaging informatics topics of image processing, visualization, and computerized detection, three new trends are emerging: (1) development of ontologies to describe radiology reports and images, (2) structured reporting and image annotation methods to make image semantics explicit and machine-accessible, and (3) applications that use semantic image information for decision support to improve radiologist interpretation performance. The informatics methods being developed have similarities and synergies with recent work in the biomedical informatics community that leverage large high-throughput data sets, and future research in imaging informatics will build on these advances to enable discovery by mining large image databases. CONCLUSIONS: Imaging informatics is beginning to develop and apply knowledge representation and analysis methods to image datasets. This type of work, already commonplace in biomedical research with large scale molecular and clinical datasets, will lead to new ways for computers to work with image data. The new advances hold promise for integrating imaging with the rest of the patient record as well as molecular data, for new data-driven discoveries in imaging analogous to that in bioinformatics, and for improved quality of radiology practice.