Machine learning-based dual-energy CT parametric mapping.

The aim is to develop and evaluate machine learning methods for generating quantitative parametric maps of effective atomic number (Zeff), relative electron density (ρ e), mean excitation energy (I x ), and relative stopping power (RSP) from clinical dual-energy CT data. The maps could be used for material identification and radiation dose calculation. Machine learning methods of historical centroid (HC), random forest (RF), and artificial neural networks (ANN) were used to learn the relationship between dual-energy CT input data and ideal output parametric maps calculated for phantoms from the known compositions of 13 tissue substitutes. After training and model selection steps, the machine learning predictors were used to generate parametric maps from independent phantom and patient input data. Precision and accuracy were evaluated using the ideal maps. This process was repeated for a range of exposure doses, and performance was compared to that of the clinically-used dual-energy, physics-based method which served as the reference. The machine learning methods generated more accurate and precise parametric maps than those obtained using the reference method. Their performance advantage was particularly evident when using data from the lowest exposure, one-fifth of a typical clinical abdomen CT acquisition. The RF method achieved the greatest accuracy. In comparison, the ANN method was only 1% less accurate but had much better computational efficiency than RF, being able to produce parametric maps in 15 s. Machine learning methods outperformed the reference method in terms of accuracy and noise tolerance when generating parametric maps, encouraging further exploration of the techniques. Among the methods we evaluated, ANN is the most suitable for clinical use due to its combination of accuracy, excellent low-noise performance, and computational efficiency.

Evaluation of hybrid SART + OS + TV iterative reconstruction algorithm for optical-CT gel dosimeter imaging.

Optical computed tomography (optical-CT) is a high-resolution, fast, and easily accessible readout modality for gel dosimeters. This paper evaluates a hybrid iterative image reconstruction algorithm for optical-CT gel dosimeter imaging, namely, the simultaneous algebraic reconstruction technique (SART) integrated with ordered subsets (OS) iteration and total variation (TV) minimization regularization. The mathematical theory and implementation workflow of the algorithm are detailed. Experiments on two different optical-CT scanners were performed for cross-platform validation. For algorithm evaluation, the iterative convergence is first shown, and peak-to-noise-ratio (PNR) and contrast-to-noise ratio (CNR) results are given with the cone-beam filtered backprojection (FDK) algorithm and the FDK results followed by median filtering (mFDK) as reference. The effect on spatial gradients and reconstruction artefacts is also investigated. The PNR curve illustrates that the results of SART + OS + TV finally converges to that of FDK but with less noise, which implies that the dose-OD calibration method for FDK is also applicable to the proposed algorithm. The CNR in selected regions-of-interest (ROIs) of SART + OS + TV results is almost double that of FDK and 50% higher than that of mFDK. The artefacts in SART + OS + TV results are still visible, but have been much suppressed with little spatial gradient loss. Based on the assessment, we can conclude that this hybrid SART + OS + TV algorithm outperforms both FDK and mFDK in denoising, preserving spatial dose gradients and reducing artefacts, and its effectiveness and efficiency are platform independent.

Ferric iron chelation lowers brain iron levels after intracerebral hemorrhage in rats but does not improve outcome.

Iron-mediated free radical damage contributes to secondary damage after intracerebral hemorrhage (ICH). Iron is released from heme after hemoglobin breakdown and accumulates in the parenchyma over days and then persists in the brain for months (e.g., hemosiderin). This non-heme iron has been linked to cerebral edema and cell death. Deferoxamine, a ferric iron chelator, has been shown to mitigate iron-mediated damage, but results vary with less protection in the collagenase model of ICH. This study used rapid-scanning X-ray fluorescence (RS-XRF), a synchrotron-based imaging technique, to spatially map total iron and other elements (zinc, calcium and sulfur) at three survival times after collagenase-induced ICH in rats. Total iron was compared to levels of non-heme iron determined by a Ferrozine-based spectrophotometry assay in separate animals. Finally, using RS-XRF we measured iron levels in ICH rats treated with deferoxamine versus saline. The non-heme iron assay showed elevations in injured striatum at 3 days and 4 weeks post-ICH, but not at 1 day. RS-XRF also detected significantly increased iron levels at comparable times, especially notable in the peri-hematoma zone. Changes in other elements were observed in some animals, but these were inconsistent among animals. Deferoxamine diminished total parenchymal iron levels but did not attenuate neurological deficits or lesion volume at 7 days. In summary, ICH significantly increased non-heme and total iron levels. We evaluated the latter and found it to be significantly lowered by deferoxamine, but its failure to attenuate injury or functional impairment in this model raises concern about successful translation to patients.

A median-Gaussian filtering framework for Moiré pattern noise removal from X-ray microscopy image.

Moiré pattern noise in Scanning Transmission X-ray Microscopy (STXM) imaging introduces significant errors in qualitative and quantitative image analysis. Due to the complex origin of the noise, it is difficult to avoid Moiré pattern noise during the image data acquisition stage. In this paper, we introduce a post-processing method for filtering Moiré pattern noise from STXM images. This method includes a semi-automatic detection of the spectral peaks in the Fourier amplitude spectrum by using a local median filter, and elimination of the spectral noise peaks using a Gaussian notch filter. The proposed median-Gaussian filtering framework shows good results for STXM images with the size of power of two, if such parameters as threshold, sizes of the median and Gaussian filters, and size of the low frequency window, have been properly selected.

How does CT image noise affect 3D deformable image registration for image-guided radiotherapy planning?

PURPOSE: To measure the sensitivity of deformable image registration to image noise. Deformable image registration can be used to map organ contours and other treatment planning data from one CT to another. These CT studies can be acquired with either conventional fan-beam CT systems or more novel cone-beam CT techniques. However, cone-beam CT images can have higher noise levels than fan-beam CT, which might reduce registration accuracy. We have investigated the effect of image quality differences on the deformable registration of fan-beam CTs and CTs with simulated cone-beam noise. METHOD: Our study used three CT studies for each of five prostate patients. Each CT was contoured by three experienced radiation oncologists. For each patient, one CT was designated the source image and the other two were target images. A deformable image registration process was used to register each source CT to each target CT and then transfer the manually drawn treatment planning contours from the source CT to the target CTs. The accuracy of the automatically transferred contours (and thus of the deformable registration process) was assessed by comparing them to the manual contours on the target CTs, with the differences evaluated with respect to interobserver variability in the manual contours. Then each of the target CTs was modified to include increased noise characteristic of cone-beam CT and the tests were repeated. Changes in registration accuracy due to increased noise were detected by monitoring changes in the automatically transferred contours. RESULTS: We found that the additional noise caused no significant loss of registration accuracy at magnitudes that exceeded what would normally be found in an actual cone-beam CT. SUMMARY: We conclude that noise levels in cone-beam CTs that might reduce manual contouring accuracy do not reduce image registration and automatic contouring accuracy.

Needle and seed segmentation in intra-operative 3D ultrasound-guided prostate brachytherapy.

In order to guide the needle to the correct location in 3D US-guided brachytherapy, the needle is continuously tracked as it is being inserted. A pre-scan before the needle insertion and a post-scan after the needle insertion are subtracted to obtain a difference image containing the needle. The image is projected along two orthogonal directions approximately perpendicular to the needle, and the 3D needle is reconstructed from the segmented needles in the two projected images. The seeds implanted with the needle are located in the cropped volume along the needle. Thus, the seeds are segmented using a tri-bar model and 3D line segment patterns. Finally, the positions of the seeds are determined using a peak detection technique. Experiments with agar and turkey/chicken phantoms as well as patient data demonstrated that our needle segmentation technique could segment the needle in near real-time with an accuracy of 0.6 mm in position and 1.0 degrees in orientation. The true-positive rate for seed segmentation is 100% for the agar phantom and 93% for the chicken phantom. The average distance to manual seed segmentation was 1.0mm for the agar phantom and 1.7 mm for the chicken phantom.

Automated localization of implanted seeds in 3D TRUS images used for prostate brachytherapy.

An algorithm has been developed in this paper to localize implanted radioactive seeds in 3D ultrasound images for a dynamic intraoperative brachytherapy procedure. Segmentation of the seeds is difficult, due to their small size in relatively low quality of transrectal ultrasound (TRUS) images. In this paper, intraoperative seed segmentation in 3D TRUS images is achieved by performing a subtraction of the image before the needle has been inserted, and the image after the seeds have been implanted. The seeds are searched in a "local" space determined by the needle position and orientation information, which are obtained from a needle segmentation algorithm. To test this approach, 3D TRUS images of the agar and chicken tissue phantoms were obtained. Within these phantoms, dummy seeds were implanted. The seed locations determined by the seed segmentation algorithm were compared with those obtained from a volumetric cone-beam flat-panel micro-CT scanner and human observers. Evaluation of the algorithm showed that the rms error in determining the seed locations using the seed segmentation algorithm was 0.98 mm in agar phantoms and 1.02 mm in chicken phantoms.

Oblique needle segmentation and tracking for 3D TRUS guided prostate brachytherapy.

An algorithm was developed in order to segment and track brachytherapy needles inserted along oblique trajectories. Three-dimensional (3D) transrectal ultrasound (TRUS) images of the rigid rod simulating the needle inserted into the tissue-mimicking agar and chicken breast phantoms were obtained to test the accuracy of the algorithm under ideal conditions. Because the robot possesses high positioning and angulation accuracies, we used the robot as a "gold standard," and compared the results of algorithm segmentation to the values measured by the robot. Our testing results showed that the accuracy of the needle segmentation algorithm depends on the needle insertion distance into the 3D TRUS image and the angulations with respect to the TRUS transducer, e.g., at a 10 degrees insertion anglulation in agar phantoms, the error of the algorithm in determining the needle tip position was less than 1 mm when the insertion distance was greater than 15 mm. Near real-time needle tracking was achieved by scanning a small volume containing the needle. Our tests also showed that, the segmentation time was less than 60 ms, and the scanning time was less than 1.2 s, when the insertion distance into the 3D TRUS image was less than 55 mm. In our needle tracking tests in chicken breast phantoms, the errors in determining the needle orientation were less than 2 degrees in robot yaw and 0.7 degrees in robot pitch orientations, for up to 20 degrees needle insertion angles with the TRUS transducer in the horizontal plane when the needle insertion distance was greater than 15 mm.

Brachytherapy needle deflection evaluation and correction.

In prostate brachytherapy, an 18-gauge needle is used to implant radioactive seeds. This thin needle can be deflected from the preplanned trajectory in the prostate, potentially resulting in a suboptimum dose pattern and at times requiring repeated needle insertion to achieve optimal dosimetry. In this paper, we report on the evaluation of brachytherapy needle deflection and bending in test phantoms and two approaches to overcome the problem. First we tested the relationship between needle deflection and insertion depth as well as whether needle bending occurred. Targeting accuracy was tested by inserting a brachytherapy needle to target 16 points in chicken tissue phantoms. By implanting dummy seeds into chicken tissue phantoms under 3D ultrasound guidance, the overall accuracy of seed implantation was determined. We evaluated methods to overcome brachytherapy needle deflection with three different insertion methods: constant orientation, constant rotation, and orientation reversal at half of the insertion depth. Our results showed that needle deflection is linear with needle insertion depth, and that no noticeable bending occurs with needle insertion into the tissue and agar phantoms. A 3D principal component analysis was performed to obtain the population distribution of needle tip and seed position relative to the target positions. Our results showed that with the constant orientation insertion method, the mean needle targeting error was 2.8 mm and the mean seed implantation error was 2.9 mm. Using the constant rotation and orientation reversal at half insertion depth methods, the deflection error was reduced. The mean needle targeting errors were 0.8 and 1.2 mm for the constant rotation and orientation reversal methods, respectively, and the seed implantation errors were 0.9 and 1.5 mm for constant rotation insertion and orientation reversal methods, respectively.

3D TRUS Image Segmentation in Prostate Brachytherapy.

Brachytherapy is a minimally invasive interventional surgery used to treat prostate cancer. It is composed of three steps: dose pre-planning, implantation of radioactive seeds, and dose post-planning. In these procedures, it is crucial to determine the positions of needles and seeds, measure the volume of the prostate gland. Three-dimensional transrectal ultrasound (TRUS) imaging has been demonstrated to be a useful technique to perform such tasks. Compared to CT, MRI or X-ray imaging, US image suffers from low contrast, image speckle and shadows, making it challenging for segmentation of needles, the prostates and seeds in the 3D TRUS images. In this paper, we reviewed 3D TRUS image segmentation methods used in prostate brachytherapy including the segmentations of the needles, the prostate, as well as the seeds. Furthermore, some experimental results with agar phantom, turkey and chicken phantom, as well as the patient data are reported.

Dynamic Intraoperative Prostate Brachytherapy Using 3D TRUS Guidance with Robot Assistance.

This paper describes a system for dynamic intraoperative prostate brachytherapy using 3D ultrasound guidance with robotic assistance. The system consists of 3D TRUS imaging, a robot and software for prostate segmentation, intraoperative planning, oblique needle segmentation and tracking, seed segmentation, and 3D dose planning. The robot and 3D TRUS coordinate systems are unified through robot and image calibrations. In 3D ultrasound images, the prostate is segmented using the discrete dynamic contour method, and optimal implantation plan is performed using geometric optimization followed by simulated annealing. The inserted needles are segmented and tracked using grey-level change in near real-time, and seed segmentation is performed using 3D line segment patterns. Needle placement accuracy of the robot at the "patient" skin was 0.15mm± 0.06mm, and needle angulation error was 0.07°. Needle targeting accuracy was 0.79mm±0.32mm.

3D TRUS guided robot assisted prostate brachytherapy.

This paper describes a system for dynamic intraoperative prostate brachytherapy using 3D ultrasound guidance with robot assistance. The system consists of 3D transrectal ultrasound (TRUS) imaging, a robot and software for prostate segmentation, 3D dose planning, oblique needle segmentation and tracking, seed segmentation, and dynamic re-planning and verification. The needle targeting accuracy of the system was 0.79 mm +/- 0.32 mm in a phantom study.

Robot-assisted 3D-TRUS guided prostate brachytherapy: system integration and validation.

Current transperineal prostate brachytherapy uses transrectal ultrasound (TRUS) guidance and a template at a fixed position to guide needles along parallel trajectories. However, pubic arch interference (PAI) with the implant path obstructs part of the prostate from being targeted by the brachytherapy needles along parallel trajectories. To solve the PAI problem, some investigators have explored other insertion trajectories than parallel, i.e., oblique. However, parallel trajectory constraints in current brachytherapy procedure do not allow oblique insertion. In this paper, we describe a robot-assisted, three-dimensional (3D) TRUS guided approach to solve this problem. Our prototype consists of a commercial robot, and a 3D TRUS imaging system including an ultrasound machine, image acquisition apparatus and 3D TRUS image reconstruction, and display software. In our approach, we use the robot as a movable needle guide, i.e., the robot positions the needle before insertion, but the physician inserts the needle into the patient's prostate. In a later phase of our work, we will include robot insertion. By unifying the robot, ultrasound transducer, and the 3D TRUS image coordinate systems, the position of the template hole can be accurately related to 3D TRUS image coordinate system, allowing accurate and consistent insertion of the needle via the template hole into the targeted position in the prostate. The unification of the various coordinate systems includes two steps, i.e., 3D image calibration and robot calibration. Our testing of the system showed that the needle placement accuracy of the robot system at the "patient's" skin position was 0.15 mm+/-0.06 mm, and the mean needle angulation error was 0.07 degrees. The fiducial localization error (FLE) in localizing the intersections of the nylon strings for image calibration was 0.13 mm, and the FLE in localizing the divots for robot calibration was 0.37 mm. The fiducial registration error for image calibration was 0.12 mm and 0.52 mm for robot calibration. The target registration error for image calibration was 0.23 mm, and 0.68 mm for robot calibration. Evaluation of the complete system showed that needles can be used to target positions in agar phantoms with a mean error of 0.79 mm+/-0.32 mm.