Evaluation of differential phase contrast cone beam CT imaging system.

Grating-based differential phase contrast (DPC) imaging enables the use of a hospital-grade X-ray tube, but compromises the image quality due to insufficiently coherent illumination. In this research, a bench-top DPC cone beam CT (DPC-CBCT) was systematically evaluated and compared with the traditional attenuation-based CBCT in terms of contrast to noise ratio, noise property, and contrast resolution through phantom studies. In order to evaluate DPC-CBCT for soft tissue imaging, breast specimen and small animal studies were carried out. Phantom studies indicate that phase image has lower-frequency noise, higher CNR, and improved contrast resolution. However, phase image quality was degraded in soft tissue imaging due to coherence loss caused by small-angle scattering. Hence dark-field imaging was introduced to quantitatively investigate small-angle scattering caused by an object. Experimental results indicate that inhomogeneous objects affect phase contrast imaging, phase image is more sensitive to noise, and its performance is material dependent. Dark-field imaging could also be used to locate and reduce phase image noise and artifact caused by small-angle scattering.

Dynamic cone beam CT angiography of carotid and cerebral arteries using canine model.

PURPOSE: This research is designed to develop and evaluate a flat-panel detector-based dynamic cone beam CT system for dynamic angiography imaging, which is able to provide both dynamic functional information and dynamic anatomic information from one multirevolution cone beam CT scan. METHODS: A dynamic cone beam CT scan acquired projections over four revolutions within a time window of 40 s after contrast agent injection through a femoral vein to cover the entire wash-in and wash-out phases. A dynamic cone beam CT reconstruction algorithm was utilized and a novel recovery method was developed to correct the time-enhancement curve of contrast flow. From the same data set, both projection-based subtraction and reconstruction-based subtraction approaches were utilized and compared to remove the background tissues and visualize the 3D vascular structure to provide the dynamic anatomic information. RESULTS: Through computer simulations, the new recovery algorithm for dynamic time-enhancement curves was optimized and showed excellent accuracy to recover the actual contrast flow. Canine model experiments also indicated that the recovered time-enhancement curves from dynamic cone beam CT imaging agreed well with that of an IV-digital subtraction angiography (DSA) study. The dynamic vascular structures reconstructed using both projection-based subtraction and reconstruction-based subtraction were almost identical as the differences between them were comparable to the background noise level. At the enhancement peak, all the major carotid and cerebral arteries and the Circle of Willis could be clearly observed. CONCLUSIONS: The proposed dynamic cone beam CT approach can accurately recover the actual contrast flow, and dynamic anatomic imaging can be obtained with high isotropic 3D resolution. This approach is promising for diagnosis and treatment planning of vascular diseases and strokes.

Enhancement of breast calcification visualization and detection using a modified PG method in Cone Beam Breast CT.

Cone Beam Breast CT is a promising diagnostic modality in breast imaging. Its isotropic 3D spatial resolution enhances the characterization of micro-calcifications in breasts that might not be easily distinguishable in mammography. However, due to dose level considerations, it is beneficial to further enhance the visualization of calcifications in Cone Beam Breast CT images that might be masked by noise. In this work, the Papoulis-Gerchberg method was modified and implemented in Cone Beam Breast CT images to improve the visualization and detectability of calcifications. First, the PG method was modified and applied to the projections acquired during the scanning process; its effects on the reconstructed images were analyzed by measuring the Modulation Transfer Function and the Noise Power Spectrum. Second, Cone Beam Breast CT images acquired at different dose levels were pre-processed using this technique to enhance the visualization of calcification. Finally, a computer-aided diagnostic algorithm was utilized to evaluate the efficacy of this method to improve calcification detectability. The results demonstrated that this technique can effectively improve image quality by improving the Modulation Transfer Function with a minor increase in noise level. Consequently, the visualization and detectability of calcifications were improved in Cone Beam Breast CT images. This technique was also proved to be useful in reducing the x-ray dose without degrading visualization and detectability of calcifications.

Scatter correction for clinical cone beam CT breast imaging based on breast phantom studies.

In flat-panel detector-based cone beam CT breast imaging (CBCTBI) systems, scattering is an important factor that degrades image quality. It is not practical to measure the scattering profiles of a breast for all view angles in a patient study, but it is possible to develop a method to estimate the scattering profiles based on information acquired from breast phantom studies. A new scattering correction method is proposed for clinical CBCTBI in this study. The scattering profiles of three anthropomorphic uncompressed breast phantoms of different sizes were thoroughly investigated, and the results indicated that though phantom size differed, the scattering profiles were mainly determined by local breast diameters, which are the approximate diameters of coronal slices that are perpendicular to the nipple-to-chestwall direction. Thus for scattering correction purposes it is possible to establish a relationship between location breast diameters and local scattering profiles, namely the fitted smooth curves of scatter-to-primary ratios (SPR) and normalized scattered radiations (NSR). In clinical CBCTBI studies, after the local breast diameters are sampled and measured on projection images, the scattering image for every projection image can be generated based on the established relationship, and the projection images can be corrected using either the SPR based method or the NSR based method. Phantom studies and clinical studies showed that both the SPR and NSR methods are able to correct cupping artifacts and reduce reconstruction error. The SPR method does not increase tissue contrast or noise while the NSR method increases both.

Cone-beam CT for breast imaging: Radiation dose, breast coverage, and image quality.

OBJECTIVE: The primary objectives of this pilot study were to evaluate the radiation dose, breast coverage, and image quality of cone-beam breast CT compared with a conventional mammographic examination. Image quality analysis was focused on the concordance of cone-beam breast CT with conventional mammography in terms of mammographic findings. SUBJECTS AND METHODS: This prospective study was performed from July 2006 through August 2008. Twenty-three women were enrolled who met the inclusion criteria, which were age 40 years or older with final BI-RADS assessment category 1 or 2 lesions on conventional mammograms within the previous 6 months. The breasts were imaged with a flat-panel detector-based cone-beam CT system, and the images were reviewed with a 3D visualization system. Cone-beam breast CT image data sets and the corresponding mammograms were reviewed by three qualified mammographers. The parameters assessed and compared in this pilot study were radiation dose, breast tissue coverage, and image quality, including detectability of masses and calcifications. The mammograms and cone-beam breast CT images were independently reviewed side by side, and the reviewers were not blinded to the other technique. The observed agreement and Cohen's kappa were used to evaluate agreement between the mammographic and cone-beam breast CT findings and interobserver agreement. Each subject responded to a questionnaire on multiple parameters, including comfort of the cone-beam breast CT examination compared with mammography. RESULTS: For a conventional mammographic examination, the average glandular radiation dose ranged from 2.2 to 15 mGy (mean, 6.5 [SD, 2.9] mGy). For cone-beam breast CT, the average glandular dose ranged from 4 to 12.8 mGy (mean, 8.2 [SD, 1.4] mGy). The average glandular dose from cone-beam breast CT was generally within the range of that from conventional mammography. For heterogeneously dense and extremely dense breasts, the difference between the mean dose of conventional mammography and that of cone-beam breast CT was not statistically significant (7.0 vs 8.1 mGy, p = 0.06). Breast tissue coverage was statistically significantly better with cone-beam breast CT than with mammography in the lateral (p < 0.0001), medial (p < 0.0001), and posterior (p = 0.0002) aspects. Mammography had statistically significantly better coverage than cone-beam breast CT in the axilla and axillary tail (p < 0.0001). Overall, most calcifications and all masses detected with mammography were also detected with cone-beam breast CT. The interobserver agreement on cone-beam breast CT was 83.7% in the detectability of imaging findings. The overall interobserver agreement on type of findings, size of findings (<1, 1-4.99, and > or = 5 mm), and location of findings was 77.2%, 84.8%, and 78.3%, respectively. CONCLUSION: The results of this study show that cone-beam breast CT can be used to image the entire breast from chest wall to nipple with sufficient spatial and contrast resolution for detection of masses and calcifications at a radiation dose within the range of that of conventional mammography.

Deep Efficient End-to-end Reconstruction (DEER) Network for Few-view Breast CT Image Reconstruction.

Breast CT provides image volumes with isotropic resolution in high contrast, enabling detection of small calcification (down to a few hundred microns in size) and subtle density differences. Since breast is sensitive to x-ray radiation, dose reduction of breast CT is an important topic, and for this purpose, few-view scanning is a main approach. In this article, we propose a Deep Efficient End-to-end Reconstruction (DEER) network for few-view breast CT image reconstruction. The major merits of our network include high dose efficiency, excellent image quality, and low model complexity. By the design, the proposed network can learn the reconstruction process with as few as O ( N ) parameters, where N is the side length of an image to be reconstructed, which represents orders of magnitude improvements relative to the state-of-the-art deep-learning-based reconstruction methods that map raw data to tomographic images directly. Also, validated on a cone-beam breast CT dataset prepared by Koning Corporation on a commercial scanner, our method demonstrates a competitive performance over the state-of-the-art reconstruction networks in terms of image quality. The source code of this paper is available at: https://github.com/HuidongXie/DEER.

Cone Beam Breast CT noise reduction using 3D adaptive Gaussian filtering.

In Cone Beam Breast CT (CBBCT) imaging, noise causes degradation of three dimensional breast images, impeding correct diagnosis of breast cancer. Within Feldkamp's cone beam reconstruction framework, applying weighted reconstruction filters to the projection images after pre-processing procedures has long been used to reduce noise and improve image quality. However, CBBCT noise is distributed across frequencies along with the useful signal. Various reconstruction filters working in the frequency domain suppress noise as well as the edge detail signal. Based on fuzzy c-means clustering and the two-dimensional histogram analysis of a large number of clinical CBBCT data, we managed to discriminate fatty stroma, glandular tissues and the transition areas between these tissues by the local mean and standard deviation values. We also proposed a three-dimensional Gaussian filtering scheme to reduce the noise in 3D reconstructed images adaptively without much blurring of detail signal.

NPS characterization and evaluation of a cone beam CT breast imaging system.

The Noise Power Spectrum (NPS) is a function that yields information about the spatial frequency composition of noise in images obtained by a system. It is evaluated by calculating the absolute value squared of the noise image and normalizing it with respect to the voxel and matrix sizes. Consequently, the NPS has been one of the physical characteristics that is commonly used to quantitatively measure the physical performance of a system. In this article, we evaluated the NPS of a Cone Beam CT Breast Imaging system by considering the following factors. First, we evaluated its symmetry around the x- and y-axis along with the influence of the cone angle and the matrix size on the NPS. Then, an analytical curve was suggested to best represent the NPS. Second, we analyzed the influence on the NPS of a set of seven parameters, namely the pixel size, exposure level, kVp value, number of projections acquired, voxel size, back projection filter, and the reconstruction algorithm employed. In addition, since the breast induced scattering in the image, we investigated the effect of the scattering-correction algorithm used in this system. Finally, we evaluated the uniformity of the NPS as a function of z with the matrix center located at {r = 0 mm}. The results demonstrate that the proposed curve is an ideal candidate that best represents the NPS. Hence, two parameters, the amplitude (A) and the width (sigma), can be used to characterize the curve. The results also demonstrate that the voxel size and the cone angle are the only two parameters investigated in this study that do not affect the NPS. On the other hand, the matrix and pixel sizes, the back-projection filter and the reconstruction algorithm, the exposure level and the scattering correction, all influence the NPS. Finally, the results of the last part of this investigation suggest that this imaging system does not have a 3D isotropic noise distribution along the z-axis; yielding less noisy images at around z = 0.00 m and z = 80 mm.

Measurements of the modulation transfer function, normalized noise power spectrum and detective quantum efficiency for two flat panel detectors: a fluoroscopic and a cone beam computer tomography flat panel detectors.

The physical performance of two Flat Panel Detectors has been evaluated. The first Flat Panel Detector is for Fluoroscopic applications, Varian PaxScan 2520, and the second is for Cone Beam Computer Tomography applications, Varian PaxScan 4030CB. First, the spectrum of the X-ray source was measured. Second, the linearity of the detectors was investigated by using an ionization chamber and the average ADU values of the detectors. Third, the temporal resolution was characterized by evaluating their image lag. Fourth, their spatial resolution was characterized by the pre-sampling Modulation Transfer Function. Fifth, the Normalized Noise Power Spectrum was calculated for various exposures levels. Finally, the Detective Quantum Efficiency was obtained as a function of spatial frequency and entrance exposure. The results illustrate that the physical performance in Detective Quantum Efficiency and Normalized Noise Power Spectrum of the Cone Beam Computer Tomography detector is superior to that of the fluoroscopic detector whereas the latter detector has a higher spatial resolution as demonstrated by larger values of its Modulation Transfer Function at large spatial frequencies.

Image denoising based on wavelets and multifractals for singularity detection.

This paper presents a very efficient algorithm for image denoising based on wavelets and multifractals for singularity detection. A challenge of image denoising is how to preserve the edges of an image when reducing noise. By modeling the intensity surface of a noisy image as statistically self-similar multifractal processes and taking advantage of the multiresolution analysis with wavelet transform to exploit the local statistical self-similarity at different scales, the pointwise singularity strength value characterizing the local singularity at each scale was calculated. By thresholding the singularity strength, wavelet coefficients at each scale were classified into two categories: the edge-related and regular wavelet coefficients and the irregular coefficients. The irregular coefficients were denoised using an approximate minimum mean-squared error (MMSE) estimation method, while the edge-related and regular wavelet coefficients were smoothed using the fuzzy weighted mean (FWM) filter aiming at preserving the edges and details when reducing noise. Furthermore, to make the FWM-based filtering more efficient for noise reduction at the lowest decomposition level, the MMSE-based filtering was performed as the first pass of denoising followed by performing the FWM-based filtering. Experimental results demonstrated that this algorithm could achieve both good visual quality and high PSNR for the denoised images.

Forest representation of vessels in cone-beam computed tomographic angiography.

Cone-beam computed tomographic angiography (CBCTA) provides a fast three-dimensional (3D) vascular imaging modality, aiming at digitally representing the spatial vascular structure in an angiographic volume. Due to the finite coverage of cone-beam scan, as well as the volume cropping in volumetric image processing, an angiographic volume may fail to contain a whole vascular tree, but rather consist of a multitude of vessel segments or subtrees. As such, it is convenient to represent multitudinal components by a forest. The vessel tracking issue then becomes component characterization/identification in the forest. The forest representation brings several conveniences for vessel tracking: (1) to sort and count the vessels in an angiographic volume, for example, according to spatial occupancy and skeleton pathlength; (2) to single out a vessel and perform in situ 3D measurement and 3D visualization in the support space; (3) to delineate individual vessels from the original angiographic volume; and (4) to cull the forest by getting rid of non-vessels and small vessels. A 3D skeletonization is used to generate component skeletons. For tree construction from skeletons, we suggest a pathlength-based procedure, which lifts the restrictions of unit-width skeleton and root determination. We experimentally demonstrate the forest representation of a dog's carotid arteries in a CBCTA system. In principle, the forest representation is useful for managing vessels in both 2D angiographic images and 3D angiographic volumes.

Cone beam breast CT with multiplanar and three dimensional visualization in differentiating breast masses compared with mammography.

OBJECTIVE: This pilot study was to evaluate cone beam breast computed tomography (CBBCT) with multiplanar and three dimensional (3D) visualization in differentiating breast masses in comparison with two-view mammograms. METHODS: Sixty-five consecutive female patients (67 breasts) were scanned by CBBCT after conventional two-view mammography (Hologic, Motarget, compression factor 0.8). For CBBCT imaging, three hundred (1024 × 768 × 16b) two-dimensional (2D) projection images were acquired by rotating the x-ray tube and a flat panel detector (FPD) 360 degree around one breast. Three-dimensional CBBCT images were reconstructed from the 2D projections. Visage CS 3.0 and Amira 5.2.2 were used to visualize reconstructed CBBCT images. RESULTS: Eighty-five breast masses in this study were evaluated and categorized under the breast imaging reporting and data system (BI-RADS) according to plain CBBCT images and two-view mammograms, respectively, prior to biopsy. BI-RADS category of each breast was compared with biopsy histopathology. The results showed that CBBCT with multiplanar and 3D visualization would be helpful to identify the margin and characteristics of breast masses. The category variance ratios for CBBCT under the BI-RADS were 23.5% for malignant tumors (MTs) and 27.3% for benign lesions in comparison with pathology, which were evidently closer to the histopathology results than those of two-view mammograms, p value <0.01. With the receiver operating characteristic (ROC) curve analysis, the area under the curve (AUC) of CBBCT was 0.911, larger than that (AUC 0.827) of two-view mammograms, p value <0.01. CONCLUSION: CBBCT will be a distinctive noninvasive technology in differentiating and categorizing breast masses under BI-RADS. CBBCT may be considerably more effective to identify breast masses, especially some small, uncertain or multifocal masses than conventional two-view mammography.

Cone-beam volume CT breast imaging: feasibility study.

X-ray projection mammography, using a film/screen combination, or digital techniques, has proven to be the most effective imaging modality currently available for early detection of breast cancer. However, the inherent superimposition of structures makes a small carcinoma (a few millimeters in size) difficult to detect when it is occult or in dense breasts, leading to a high false-positive biopsy rate. Cone-beam x-ray-projection-based volume imaging using flat panel detectors (FPDs) may allow obtaining three-dimensional breast images, resulting in more accurate diagnosis of structures and patterns of lesions while eliminating the hard compression of breasts. This article presents a novel cone-beam volume computed tomographic breast imaging (CBVCTBI) technique based on the above techniques. Through a variety of computer simulations, the key issues of the system and imaging techniques were addressed, including the x-ray imaging geometry and corresponding reconstruction algorithms, x-ray characteristics of breast tissue and lesions, x-ray setting techniques, the absorbed dose estimation, and the quantitative effect of x-ray scattering on image quality. The preliminary simulation results support the proposed CVBCTBI modality for breast imaging in respect to its feasibility and practicability. The absorbed dose level is comparable to that of current mammography and will not be a prominent problem for this imaging technique. Compared to conventional mammography, the proposed imaging technique with isotropic spatial resolution will potentially provide significantly better low-contrast detectability of breast tumors and more accurate location of breast lesions.

X-ray scatter correction algorithm for cone beam CT imaging.

Developing and optimizing an x-ray scatter control and reduction technique is one of the major challenges for cone beam computed tomography (CBCT) because CBCT will be much less immune to scatter than fan-beam CT. X-ray scatter reduces image contrast, increases image noise and introduces reconstruction error into CBCT. To reduce scatter interference, a practical algorithm that is based upon the beam stop array technique and image sequence processing has been developed on a flat panel detector-based CBCT prototype scanner. This paper presents a beam stop array-based scatter correction algorithm and the evaluation results through phantom studies. The results indicate that the beam stop array-based scatter correction algorithm is practical and effective to reduce and correct x-ray scatter for a CBCT imaging task.

Flat panel detector-based cone beam computed tomography with a circle-plus-two-arcs data acquisition orbit: preliminary phantom study.

Cone beam computed tomography (CBCT) has been investigated in the past two decades due to its potential advantages over a fan beam CT. These advantages include (a) great improvement in data acquisition efficiency, spatial resolution, and spatial resolution uniformity, (b) substantially better utilization of x-ray photons generated by the x-ray tube compared to a fan beam CT, and (c) significant advancement in clinical three-dimensional (3D) CT applications. However, most studies of CBCT in the past are focused on cone beam data acquisition theories and reconstruction algorithms. The recent development of x-ray flat panel detectors (FPD) has made CBCT imaging feasible and practical. This paper reports a newly built flat panel detector-based CBCT prototype scanner and presents the results of the preliminary evaluation of the prototype through a phantom study. The prototype consisted of an x-ray tube, a flat panel detector, a GE 8800 CT gantry, a patient table and a computer system. The prototype was constructed by modifying a GE 8800 CT gantry such that both a single-circle cone beam acquisition orbit and a circle-plus-two-arcs orbit can be achieved. With a circle-plus-two-arcs orbit, a complete set of cone beam projection data can be obtained, consisting of a set of circle projections and a set of arc projections. Using the prototype scanner, the set of circle projections were acquired by rotating the x-ray tube and the FPD together on the gantry, and the set of arc projections were obtained by tilting the gantry while the x-ray tube and detector were at the 12 and 6 o'clock positions, respectively. A filtered backprojection exact cone beam reconstruction algorithm based on a circle-plus-two-arcs orbit was used for cone beam reconstruction from both the circle and arc projections. The system was first characterized in terms of the linearity and dynamic range of the detector. Then the uniformity, spatial resolution and low contrast resolution were assessed using different phantoms mainly in the central plane of the cone beam reconstruction. Finally, the reconstruction accuracy of using the circle-plus-two-arcs orbit and its related filtered backprojection cone beam volume CT reconstruction algorithm was evaluated with a specially designed disk phantom. The results obtained using the new cone beam acquisition orbit and the related reconstruction algorithm were compared to those obtained using a single-circle cone beam geometry and Feldkamp's algorithm in terms of reconstruction accuracy. The results of the study demonstrate that the circle-plus-two-arcs cone beam orbit is achievable in practice. Also, the reconstruction accuracy of cone beam reconstruction is significantly improved with the circle-plus-two-arcs orbit and its related exact CB-FPB algorithm, as compared to using a single circle cone beam orbit and Feldkamp's algorithm.

Image denoising based on multiscale singularity detection for cone beam CT breast imaging.

It was recently reported that the real-time flat panel detector-based cone-beam computed tomography (CBCT) breast imaging can help improve the detectability of small breast tumors with an X-ray dose comparable to that of the conventional mammography. In this paper, an efficient denoising algorithm is proposed to further reduce the X-ray exposure level required by a CBCT scan to acquire acceptable image quality. The proposed wavelet-based denoising algorithm possesses three significant characteristics: 1) wavelet coefficients at each scale are classified into two categories: irregular coefficients, and edge-related and regular coefficients; 2) noise in irregular coefficients is reduced as much as possible without producing artifacts to the denoised images; and 3) for the edge-related and regular coefficients, if they are at the first decomposition level, they are further denoised, otherwise, no modifications are made to them so as to obtain good visual quality for diagnosis. By applying the proposed denoising algorithm to the filtered projection images, the X-ray exposure level necessary for the CBCT scan can be reduced by up to 60% while obtaining clinically acceptable image quality. This denoising result indicates that in the clinical application of CBCT breast imaging, the patient radiation dose can be significantly reduced.

Three-dimensional point spread function measurement of cone-beam computed tomography system by iterative edge-blurring algorithm.

With separability assumed, we decompose a three-dimensional point spread function (3D PSF) into two-dimensional (2D) PSFs and further into one-dimensional (ID) PSFs. Based on the observation of the location invariance of a step edge under convolution, we propose a rectification procedure to automatically establish the step-edge function from a blurred edge profile. The ID PSF is modelled as a single-parameter Gaussian function, which is determined by iteratively blurring a step-edge function into a spread edge profile. A plastic solid ball (diameter approximately 6 mm) is used to provide double-edged rectangular functions along scanlines passing through the ball centre, and correspondingly, the reconstructed digital volume provides the blurred rectangular profiles. Experimenting with a cone-beam computed tomography system, we demonstrate the iterative edge-blurring algorithm for PSF measurement. By repositioning the ball phantom in the object support space, we measure the system's spatial variance in terms of full-width-at-half-maximum (FWHM) of the local PSFs. Specifically, we obtained the FWHMs for three specific locations at (0, 0, -40 mm), (0, 0, 0) and (0, 0, 40 mm), which are given by 0.92 +/- 0.10 mm, 0.65 +/- 0.08 mm and 0.93 +/- 0.10 mm, respectively.

Why should breast tumour detection go three dimensional?

Although x-ray mammography is widely developed for breast tumour detection, it suffers from spatial superposition in its two-dimensional (2D) representation of a three-dimensional (3D) breast structure. Accordingly, 3D breast imaging, such as cone-beam computed tomography (CT), arises at the historic moment. In this paper, we theoretically elucidate the spatial superposition effect associated with x-ray mammography on breast tumour detection. This explanation is based on the line integral of x-ray traversing a composite breast model. As a result, we can characterize the difficulty of detecting small tumours in terms of local intensity contrast in x-ray images. In comparison, we also introduce cone-beam CT breast imaging for 3D breast volume representation, which offers advantages for breast mass segmentation and measurement. The discussion is demonstrated with an experiment with a breast surgical specimen. In conclusion, we strongly believe that 3D volumetric representation allows for more accurate breast tumour detection.

Breast volume denoising and noise characterization by 3D wavelet transform.

Breast imaging through cone-beam computed tomography provides a digital breast volume, with which the three-dimensional (3D) breast tissues can be analyzed. Data denoising, as a preprocessing step for subsequent volumetric breast segmentation is always needed. In this paper, we report a volumetric denoising technique by a separable 3D wavelet transform (WT), i.e. a '2D WT plus 1D WT' scheme. Specifically, the scheme performs two-dimensional (2D) wavelet denoising on a stack of slice images of the breast volume, followed by one-dimensional (1D) wavelet denoising along the stacking direction. The denoising is achieved by wavelet decomposition, high-pass subband attenuation, and wavelet synthesis. A one-level 3D WT produces eight subbands occupying the octants of the 3D wavelet space. Multilevel WT also provides a multiresolution representation of breast volume, i.e. a sequence of low-pass subbands. In general, most noise and irregularity features are imparted into the high-pass subbands, which are removed or reduced for denoising purpose. Meanwhile, the information in a subband can be characterized in terms of energy, variance, and entropy. Through 3D visualization, the spatial structure in a subband can also be visually perceived. Experimental demonstration with the breast volume reconstructed from a specimen is provided.

Investigating the use of texture features for analysis of breast lesions on contrast-enhanced cone beam CT.

Cone beam computed tomography (CBCT) has found use in mammography for imaging the entire breast with sufficient spatial resolution at a radiation dose within the range of that of conventional mammography. Recently, enhancement of lesion tissue through the use of contrast agents has been proposed for cone beam CT. This study investigates whether the use of such contrast agents improves the ability of texture features to differentiate lesion texture from healthy tissue on CBCT in an automated manner. For this purpose, 9 lesions were annotated by an experienced radiologist on both regular and contrast-enhanced CBCT images using two-dimensional (2D) square ROIs. These lesions were then segmented, and each pixel within the lesion ROI was assigned a label - lesion or non-lesion, based on the segmentation mask. On both sets of CBCT images, four three-dimensional (3D) Minkowski Functionals were used to characterize the local topology at each pixel. The resulting feature vectors were then used in a machine learning task involving support vector regression with a linear kernel (SVRlin) to classify each pixel as belonging to the lesion or non-lesion region of the ROI. Classification performance was assessed using the area under the receiver-operating characteristic (ROC) curve (AUC). Minkowski Functionals derived from contrast-enhanced CBCT images were found to exhibit significantly better performance at distinguishing between lesion and non-lesion areas within the ROI when compared to those extracted from CBCT images without contrast enhancement (p < 0.05). Thus, contrast enhancement in CBCT can improve the ability of texture features to distinguish lesions from surrounding healthy tissue.