CT Images Of Phantom Research Articles

Evaluation of direct point dose estimation based on the distribution of the size-specific dose estimate.

The aim of this study was to evaluate the point doses using a distribution of the size-specific dose estimate (SSDE) from axial CT images of in-house phantoms having diameters from 8 to 40cm. In-house phantoms made of polyester-resin (PESR) mixed with methyl ethyl ketone peroxide (MEKP) were used. The phantoms were built with different diameter sizes of 8, 16, 24, 32, and 40cm. The phantoms were scanned by Siemens a SOMATOM Perspective-128 slice CT scanner with constant input parameters. The point doses were interpolated from the central SSDE (SSDEc) and the peripheral SSDE (SSDEp). The SSDEc and SSDEp were calculated from the SSDE with h- and k-factors. The point doses were compared to the direct measurements using the nanoDot™ optically-stimulated luminescence dosimeter (OSLD) in dedicated holes on the phantoms. It was found that the point dose decreases as the phantom diameter increased. The doses obtained using two approaches differed by 11% on average. The highest difference was 40% and the lowest difference was < 1%. It was found that dose based on the SSDE concept tended to be higher compared to the measured dose by OSLD. Point dose estimation using the concept of SSDE distribution can be considered an alternative for accurate and simple estimation. This approach still requires improvements to increase its accuracy and its application to estimate the organ dose needs further investigation.

A novel AP/PA total body irradiation technique using abutting IMRT fields at extended SSD.

To develop a Total Body Irradiation (TBI) technique using IMRT at extended SSD that can be performed in any size Linac room. Patients studied were placed on a platform close to the floor, directly under the gantry with cranial-caudal axis parallel to the gantry rotation plane and at SSD ∼200cm. Two abutting fields with the same external isocenter at gantry angles of±21˚, collimator angle of 90˚, and field size of 25×40 cm2 are employed for both supine and prone positions. An iterative optimization algorithm was developed to generate a uniform dose at the patient mid-plane with adequate shielding to critical organs such as lungs and kidneys. The technique was validated in both phantom and patient CT images for treatment planning, and dose measurement and QA were performed in phantom. A uniform dose distribution in the mid-plane within±5% of the prescription dose was reached after a few iterations. This was confirmed with ion-chamber measurements in phantom. The mean dose to lungs and kidneys can be adjusted according to clinical requirements and can be as low as ∼25% of the prescription dose. For a typical prescription dose of 200 cGy/fraction, the total MU was ∼2400/1200 for the superior/inferior field. The overall treatment time for both supine/prone positions was ∼54 min to meet the maximum absorbed dose rate criteria of 15 cGy/min. IMRT QA with portal dosimetry shows excellent agreement. We have developed a promising TBI technique using abutting IMRT fields at extended SSD. The patient is in a comfortable recumbent position with good reproducibility and less motion during treatment. An additional benefit of this technique is that full 3D dose distribution is available from the TPS with a DVH summary for organs of interest. The technique allows precise sparing of lungs and kidneys and can be executed in any linac room.

Distance between Slices Investigation on Siemens CT Phantom Images

This study aims to develop an algorithm for automatic measurement of distance between slices on a Siemens CT phantom with variations of slice thickness and field of view (FOV). The Siemens 64-slice Somatom Perspective CT scanner was utilized to acquire the Siemens phantom image data, with slice thickness variation (i.e., 1, 2, 3, 4, and 5 mm) and FOV variation (i.e., 220, 240, and 300 mm). The distance between slices was calculated by comparing two images from different slice positions. The first stage was opening the images. The second stage was segmentation of the ramp object from image slice-1. The third stage was determination of the centroid coordinates of the ramp object from image slice-1. The fourth stage repeated the processes at second and third stages for slice-2. The fifth stage was calculating the distance between slices. The measured distances were compared with the set distances extracting from DICOM header for every slice position. The developed algorithm can measure distance between slices for variations of slice thickness and FOV. The results of automatic measurement of distance between slices have strong linear correlation ( &gt; 0.99) with the set distances. The differences between automated measurement results and set distances are more than 2 mm for distances up to 50 mm or the differences are around 23% for both slice thickness and FOV variations.

Comparison of MTFs Measured using IndoQCT and ImQuest Software on GE CT Phantom Images

Purpose: This study aims to compare the modulation transfer function (MTF) measured using IndoQCT and ImQuest software on GE CT phantom images with variations of slice thickness and reconstruction filter. Method: This study compared MTFs measured using two software (i.e., IndoQCT and ImQuest software) on GE CT phantom images. IndoQCT implemented automatic region of interest (ROI) determination, while ImQuest used manual ROI determination. Both software analysed images with three variations of slice thickness (i.e., 2.5, 5, and 10 mm) and reconstruction filter (i.e., standard, soft tissue, chest, and bone filters). Results: IndoQCT provides automatic ROI selection and produces stable MTF curves, while ImQuest provides manual ROI selection and produces more fluctuating MTF curves. In slice thickness variation, IndoQCT showed that different slice thickness did not significantly affect spatial resolution as suggested, while ImQuest showed significant differences in spatial resolution for thickest slice thickness. In filter variation, IndoQCT produced high spatial resolution value in bone filter, while ImQuest produced the lower spatial resolution value. Overall, both software showed MTF differences within 15%, except in certain conditions, for example on a slice thickness of 10 mm. Conclusions: We have compared MTFs measured using IndoQCT and ImQuest software. Both software can be used to measure MTFs on the GE phantom images with relatively accurate results. IndoQCT generally produces higher MTF values than those from ImQuest. However, the difference MTF is not statistically significant. This study found that at 10 mm slice thickness, ImQuest produced very low MTF, while IndoQCT produced stable results for different slice thicknesses and reconstruction filters.

Automatic slice thickness measurement on three types of Catphan CT phantoms

Objective. To develop an algorithm to measure slice thickness running on three types of Catphan phantoms with the ability to adapt to any misalignment and rotation of the phantoms. Method. Images of Catphan 500, 504, and 604 phantoms were examined. In addition, images with various slice thicknesses ranging from 1.5 to 10.0 mm, distance to the iso-center and phantom rotations were also examined. The automatic slice thickness algorithm was carried out by processing only objects within a circle having a diameter of half the diameter of the phantom. A segmentation was performed within an inner circle with dynamic thresholds to produce binary images with wire and bead objects within it. Region properties were used to distinguish wire ramps and bead objects. At each identified wire ramp, the angle was detected using the Hough transform. Profile lines were then placed on each ramp based on the centroid coordinates and detected angles, and the full-width at half maximum (FWHM) was determined for the average profile. The slice thickness was obtained by multiplying the FWHM by the tangent of the ramp angle (23°). Results. Automatic measurements work well and have only a small difference (<0.5 mm) from manual measurements. For slice thickness variation, automatic measurement successfully performs segmentation and correctly locates the profile line on all wire ramps. The results show measured slice thicknesses that are close (<3 mm) to the nominal thickness at thin slices, but slightly deviated for thicker slices. There is a strong correlation (R2 = 0.873) between automatic and manual measurements. Testing the algorithm at various distances from the iso-center and phantom rotation angle also produced accurate results. Conclusion. An automated algorithm for measuring slice thickness on three types of Catphan CT phantom images has been developed. The algorithm works well on various thicknesses, distances from the iso-center, and phantom rotations.

UNet and MobileNet CNN-based model observers for CT protocol optimization: comparative performance evaluation by means of phantom CT images.

The aim of this work is the development and characterization of a model observer (MO) based on convolutional neural networks (CNNs), trained to mimic human observers in image evaluation in terms of detection and localization of low-contrast objects in CT scans acquired on a reference phantom. The final goal is automatic image quality evaluation and CT protocol optimization to fulfill the ALARA principle. Preliminary work was carried out to collect localization confidence ratings of human observers for signal presence/absence from a dataset of 30,000 CT images acquired on a PolyMethyl MethAcrylate phantom containing inserts filled with iodinated contrast media at different concentrations. The collected data were used to generate the labels for the training of the artificial neural networks. We developed and compared two CNN architectures based respectively on Unet and MobileNetV2, specifically adapted to achieve the double tasks of classification and localization. The CNN evaluation was performed by computing the area under localization-ROC curve (LAUC) and accuracy metrics on the test dataset. The mean of absolute percentage error between the LAUC of the human observer and MO was found to be below 5% for the most significative test data subsets. An elevated inter-rater agreement was achieved in terms of S-statistics and other common statistical indices. Very good agreement was measured between the human observer and MO, as well as between the performance of the two algorithms. Therefore, this work is highly supportive of the feasibility of employing CNN-MO combined with a specifically designed phantom for CT protocol optimization programs.

Evaluation of the accuracy of a six-degree-of-freedom robotic couch using optical surface and cone beam CT images of an SRS QA phantom

Abstract Purpose: To assess the accuracy of the Varian PerfectPitch six-degree-of-freedom (6DOF) robotic couch by using a Varian SRS QA phantom. Methods: The stereotactic radiosurgery (SRS) phantom has five tungsten carbide BBs each with 7·5 mm in diameter arranged with the known geometry. Optical surface images and cone beam CT (CBCT) images of the phantom were taken at different pitch, roll and rotation angles. The pitch, roll, and rotation angles were varied from −3 to 3 degrees by inputs from the linac console. A total of 39 Vision RT images with different rotation angle combinations were collected, and the Vision RT software was used to determine the rotation angles and translational shifts from those images. Eight CBCT images at most allowed rotational angles were analysed by in-house software. The software took the coordinates of the voxel of the maximum CT number inside a 7·5-mm sphere surrounding one BB to be the measured position of this BB. Expected BB positions at different rotation angles were determined by multiplying measured BB positions at zero pitch and roll values by a rotation matrix. Applying the rotation matrix to 5 BB positions yielded 15 equations. A linear least square method was used for regression analysis to approximate the solutions of those equations. Results: Of the eight calculations from CBCT images, the maximum rotation angle differences (degree) were 0·10 for pitch, 0·15 for roll and 0·09 for yaw. The maximum translation differences were 0·3 mm in the left–right direction, 0·5 mm in the anterior–posterior direction and 0·4 mm in the superior–inferior direction. Conclusions: The uncertainties of the 6-DOF couch were examined with the methods of optical surface imaging and CBCT imaging of the SRS QA phantom. The rotational errors were less than 0·2 degree, and the isocentre shifts were less than 0·8 mm.

ANALYSIS OF THE EFFECT OF THE X-RAY TUBE CURRENT ON UNIFORMITY IMAGE NOISE VALUES

This research aimed to evaluate the effects of the X-ray current tube on the uniformity value of CT image noise. This research used the CT images of Siemens head water phantom. The phantom was scanned with a Siemens Somatom Scope CT scanner for various current tubes (i.e., 180 mAs, 200 mAs, 220 mAs, 240 mAs, and 260 mAs) on tube voltage 110 kV, slice thickness 3 mm, and FOV 240 mm. The uniformity values in CT image noise were analyzed using the standard deviation value (SD). The SD was measured by the ROI process in 5 different locations, in the center and at 3 o'clock, 6 o'clock, 9 o'clock, and 12 o'clock on the images CT. Based on the result of a simple linear regression test using IBM SPSS ver 25, it was found that the current tube significantly affected the uniformity value of image noise (r2=0,9768, p&lt;0,05). This result showed that increasing the current tube can reduce image noise uniformity. The highest image noise uniformity value was at a current at 180 mAs with 0,217 HU, and the lowest was at 260 mAs with 0,031 HU. This result also shows that all the values are less than 2 HU and still within the acceptable limit by the BAPETEN standard regulation.

A novel edge gradient distance metric for automated evaluation of deformable image registration quality

PurposeTo develop a new registration quality metric, based on the distance between image edges, for automated evaluation and comparison of DIR algorithms. MethodsCanny filter is used to create binary gradient images from input images to be compared. A small subregion of one binary image is translated relative to the other. The translational distance maximizing overlap of edges in the subregion is the local edge gradient distance to agreement (EGDTA); repeating over all subregions provides an EGDTA map. The method was tested on phantom and pelvic CT images, by applying a known deformable vector field (DVF). The method was then applied to evaluate two DIR algorithms (SICLE and Demons) for pelvic CTs from five patients. Three SICLE variants were used: Grayscale-driven (G), Contour-driven (C), and Grayscale + Contour-driven (G + C). For each patient, a planning CT was registered to three repeat CTs using the above DIR algorithms. Mean EGDTA values in concentric ring regions of interest close to and far away from the pelvic organ contours were compared. ResultsEGDTA maps and imposed DVF deformations on phantom and CT images demonstrated agreement. In comparison of the three variants of SICLE: C had lower EGDTA values close to the pelvic organs, while G showed much better performance in the regions distant from the organs compared to C; and G + C registration exhibited the lowest or comparable EGDTA value among three variants. Demons achieved the lowest EGDTA values. ConclusionsThe EGDTA metric shows potential as an automated means of evaluating and comparing DIR algorithms.

IQMetrix-CT: New software for task-based image quality assessment of phantom CT images

The purpose of this article was to explain how the new iQMetrix-CT software works, as well as to describe its current potential and discuss its future applications. iQMetrix-CT was developed by a working group from the French Society of Medical Physicists (SFPM). This software calculates three advanced metrics, which have been adapted to take into account the non-linear and non-stationary properties of iterative reconstruction algorithms. Noise power spectrum is calculated to assess noise texture and noise magnitude in the frequency domain. The task-based transfer function is computed to assess the spatial resolution under conditions of contrast and noise similar to the lesions encountered in clinical practice. Finally, the detectability index is used to estimate the radiologist's ability to perform a given task such as detecting a simulated lesion. These metrics are very useful to evaluate the performance of reconstruction algorithms in term of image quality and optimize the doses of a given protocol or to evaluate and compare a new technology.

Evaluation of automatic tube current modulation of CT scanners using a dedicated and the CTDI dosimetry phantoms.

PurposeTo investigate the operation principles of the automatic tube current modulation (ATCM) of a CT scanner, using a dedicated phantom and the CT dosimetry index (CTDI) phantom.Material and methodsThe Mercury 4.0 phantom and three different configurations of the CTDI dosimetry phantom were employed. A frequently used clinical scanning protocol was employed as a basis for the acquisitions performed with all phantoms, using both scanning directions. Additional acquisitions with different pitch and examination protocols were performed with Mercury phantom, to further explore their effect on ATCM and the resulting image quality. Different software named DICOM Info Extractor, ImageJ, and imQuest, were used to derive CTDIvol and table position, image noise, and water equivalent diameter (WED) of each phantom CT image, respectively. ImQuest was also used to derive the detectability index (d’) of five different materials (air, solid water, polystyrene, iodine, and bone) embedded in the Mercury phantom.ResultsIt was exhibited with all four phantoms that the scanning direction greatly affects the modulation curves. The fitting of the dose modulations curves suggested that for each table position what determines the CTDIvol value is the WED values of the phantom structures laying ahead towards the scanning direction, for a length equal to the effective width of the X‐ray beam. Furthermore, it was also exhibited that ATCM does not fully compensate for larger thicknesses, since images of larger WED phantom sections present more noise (larger SD) in all four phantoms and in Mercury 4.0 phantom smaller detectability (d’).ConclusionMercury 4.0 is a dedicated phantom for a complete and in‐depth evaluation of the ATCM operation and the resulting image quality. However, in its absence, different CTDI configurations can be used as an alternative to investigate and comprehend some basic operation principles of the CT scanners’ ATCM systems.

Automatic head computed tomography image noise quantification with deep learning

PurposeComputed tomography (CT) image noise is usually determined by standard deviation (SD) of pixel values from uniform image regions. This study investigates how deep learning (DL) could be applied in head CT image noise estimation. MethodsTwo approaches were investigated for noise image estimation of a single acquisition image: direct noise image estimation using supervised DnCNN convolutional neural network (CNN) architecture, and subtraction of a denoised image estimated with denoising UNet-CNN experimented with supervised and unsupervised noise2noise training approaches. Noise was assessed with local SD maps using 3D- and 2D-CNN architectures. Anthropomorphic phantom CT image dataset (N = 9 scans, 3 repetitions) was used for DL-model comparisons. Mean square error (MSE) and mean absolute percentage errors (MAPE) of SD values were determined using the SD values of subtraction images as ground truth. Open-source clinical head CT low-dose dataset (Ntrain = 37, Ntest = 10 subjects) were used to demonstrate DL applicability in noise estimation from manually labeled uniform regions and in automated noise and contrast assessment. ResultsThe direct SD estimation using 3D-CNN was the most accurate assessment method when comparing in phantom dataset (MAPE = 15.5%, MSE = 6.3HU). Unsupervised noise2noise approach provided only slightly inferior results (MAPE = 20.2%, MSE = 13.7HU). 2DCNN and unsupervised UNet models provided the smallest MSE on clinical labeled uniform regions. ConclusionsDL-based clinical image assessment is feasible and provides acceptable accuracy as compared to true image noise. Noise2noise approach may be feasible in clinical use where no ground truth data is available. Noise estimation combined with tissue segmentation may enable more comprehensive image quality characterization.

Treatment planning system commissioning of the first clinical biology-guided radiotherapy machine.

PurposeThe RefleXion X1 is a novel radiotherapy machine designed for image‐guided radiotherapy (IGRT) and biology‐guided radiotherapy (BgRT). Its treatment planning system (TPS) generates IMRT and SBRT plans for a 6MV‐FFF beam delivered axially via 50 firing positions with the couch advancing every 2.1 mm. The purpose of this work is to report the TPS commissioning results for the first clinical installation of RefleXion™ X1.MethodsCT images of multiple phantoms were imported into the RefleXion TPS to evaluate the accuracy of data transfer, anatomical modeling, plan evaluation, and dose calculation. Comparisons were made between the X1, Eclipse™, and MIM™. Dosimetric parameters for open static fields were evaluated in water and heterogeneous slab phantoms. Representative clinical IMRT and SBRT cases were planned and verified with ion chamber, film, and ArcCHECK@ measurements. The agreement between TPS and measurements for various clinical plans was evaluated using Gamma analysis with a criterion of 3%/2 mm for ArcCHECK@ and film. End‐to‐end (E2E) testing was performed using anthropomorphic head and lung phantoms.ResultsThe average difference between the TPS‐reported and known HU values was −1.4 ± 6.0 HU. For static fields, the agreements between the TPS‐calculated and measured PDD10, crossline profiles, and inline profiles (FWHM) were within 1.5%, 1.3%, and 0.5 mm, respectively. Measured output factors agreed with the TPS within 1.3%. Measured and calculated dose for static fields in heterogeneous phantoms agreed within 2.5%. The ArcCHECK@ mean absolute Gamma passing rate was 96.4% ± 3.4% for TG 119 and TG 244 plans and 97.8% ± 3.6% for the 21 clinical plans. E2E film analysis showed 0.8 mm total targeting error for isocentric and 1.1 mm for off‐axis treatments.ConclusionsThe TPS commissioning results of the RefleXion X1 TPS were within the tolerances specified by AAPM TG 53, MPPG 5.a, TG 119, and TG 148. A subset of the commissioning tests has been identified as baseline data for an ongoing QA program.

Impact of pixel value truncation on image quality of low dose chest CT.

In some noisy low dose CT lung cancer screening images, we noticed that the CT density values of air were increased and the visibility of emphysema was distinctly decreased. By examining histograms of these images, we found that the CT density values were truncated at -1024 HU. The purpose of this study was to investigate the effect of pixel value truncation on the visibility of emphysema using mathematical models. Assuming CT noise follows a normal distribution, we derived the relationship between the mean CT density value and the standard deviation (SD) when the pixel values below -1024 HU are truncated and replaced by -1024 HU. To validate our mathematical model, 20 untruncated phantom CT images were truncated by simulation, and the mean CT density values and SD of air in the images were measured and compared with the theoretical values. In addition, the mean CT density values and SD of air were measured in 100 cases of real clinical images obtained by GE, Siemens, and Philips scanners, respectively, and the agreement with the theoretical values was examined. Next, the contrast-to-noise ratio (CNR) between air (-1000 HU) and lung parenchyma (-850 HU) was derived from the mathematical model in the presence and absence of truncation as a measure of the visibility of emphysema. In addition, the radiation dose ratios required to obtain the same CNR in the case with and without truncation were also calculated. The mathematical model revealed that when the pixel values are truncated, the mean CT density values are proportional to the noise magnitude when the magnitude exceeds a certain level. The mean CT density values and SD measured in the images with pixel values truncated by simulation and in the real clinical images acquired by GE and Philips scanners agreed well with the theoretical values from our mathematical model. In the Siemens images, the measured and theoretical values agreed well when a portion of the truncated values were replaced by random values instead of simply replacing by -1024 HU. The CNR of air and lung parenchyma was lowered by truncating CT density values compared to that of no truncation. Furthermore, it was found that higher radiation dose was required to obtain the same CNR with truncation as without. As an example, when the noise SD was 60 HU, the radiation dose required for the GE and Philips truncation method was about 1.2 times higher than that without truncation, and that for the Siemens truncation method was about 1.4 times higher. It was demonstrated mathematically that pixel value truncation causes a brightening of the mean CT density value and decreases the CNR of emphysema. Our results indicate that it is advisable to turn off truncation at -1024 HU, especially when scanning at low and ultra-low radiation doses in the thorax.

Iterative CT reconstruction based on ADMM using shearlet sparse regularization.

The total variation (TV) method favors solutions with the piece-wise constant assumption of the desired image from sparse-view sampling, for example, simple geometric images with flat intensity. When the phantoms become more complex and contain complicated textures, for example, high-resolution phantom and lung CT images, the images reconstructed by TV regularization may lose their contrast and fine structures. One of the optimally sparse transforms for images, the shearlet transform, has C2 without discontinuities on C2 curves, giving excellent sensitive directional information as compared with other wavelet transform approaches. Here, we developed a Shearlet-Sparse Regularization (SSR) algorithm solved with the Alternating Direction Method of Multipliers (ADMM) to overcome this limitation. With the strengthened characteristics of SSR, we performed one simulation experiment and two real experiments using a NeuViz 64 X-ray CT scanning system to measure the performance and properties of proposed algorithm. The results demonstrate that the SSR method exhibits the advantage of providing high-quality directional information and contrast as compared with TV.

Dose distrubution evaluation of different dental implants on a real human dry-skull model for head and neck cancer radiotherapy

ObjectivesIn this study, the effect of different dental implant materials on dose distribution in head and neck cancer radiotherapy planning was investigated in a real human dry-skull taken from cadaver. MethodsThe effect of Root-shaped Yttrium (Y) stabilized tetragonal Zr polycrystalline (Y-TZP), Titanium (Ti), Alumina (Al2O3) and polyether ether ketone (PEEK) dental implants on dose distribution was investigated by three different methods; Monte Carlo (MC) simulation, Eclipse Treatment Planning System's (TPS) Pencil Beam Convolution (PBC) algorithm and Thermoluminescence Dosimeter (TLD). The 6 MV photon beam simulation was performed for the Varian 2300 C/D linear accelerator using the EGSnrc based BEAMnrc MC Code. To perform MC simulation on skull phantom, phantom CT images with extended CT scale were imported to BEAMnrc/EGSnrc MC code via CTCREATE code. ResultsAccording to the dose distribution measurement in dry skull, the calculated dose differences between PBC and MC algorithms were in the range of 0.6–4.5% whereas the dose difference between MC and TLD were in the range of 1,1–14.6% range depending on implant material. The PBC calculated dose increase difference range confirmed with TLD measurements was between 1.7% and 19.8%. The PBC algorithm could not fully determine the dose due to backscattering caused by dental implants. This is probably due to the limitations and shortcomings of the PBC algorithm. ConclusionThe maximum dose increase was observed for the Y-TZP dental implant material, which has the highest physical density. No significant dose increase was observed for the PEEK implant probably due to the low physical density of implants close to the physical density of water.

Experimental determination of Gd dose enhancement and Gd dose sparing by 192Ir brachytherapy source with Gafchromic EBT3 dosimeter

This work evaluates experimentally the dose enhancement factor (DEF) and dose sparing factor (DSF) due to radiation self-shielding, produced by Gd infused in tumor phantom irradiated with brachytherapy HDR 192Ir source by Gafchromic EBT3 dosimeter. The phantom was made of a set of solid water slabs (30 × 30 × 1.0) cm3 and three acrylic slabs of (30 × 30 × 0.5) cm3 machined to contain in the central axis acrylics vials of (1 × 1 × 5) cm3. The first and second acrylic vials were filled with an identical Gd solution of 0, 10 and 20 mg/ml, simulating Gd-doped and undoped tumor, and the third vial was filled in all the measurement only with water, representing an organ at risk. Additional solid water slabs were used to complete a phantom of (30 × 30 × 16) cm3. In the phantom center an acrylic slab was machined to introduce the 2.5 mm flexible guide tube of GammaMed plus iX equipment and positioning the 192Ir source in the phantom central part. EBT3 fragments of (0.9 × 4) cm2 were placed on the inner edge of the second and third vials to measure dose enhancement and dose sparing simultaneously. Phantom CT images were acquired for planning and to prescribe a dose of 6.0 Gy at 2.0 cm of the source, achieving an isodose curve of 44.5% at 3.0 cm (positions of the EBT3 films). Additionally, Monte Carlo simulation of the identical experimental setup was implemented to compare measurement values. The results showed the feasibility of measuring a DEF of 1.15 ± 0.05 in 20 mg/ml of Gd concentration consistent with the Monte Carlo DEF of 1.112 ± 0.005 for the same concentration. DEF value for concentration of 10 mg/ml would not be detected (1.00 ± 0.04) by an expected under measurement of the EBT3 films associated with the non-detection of photoelectrons and Auger electrons of very low energy that cannot reach the radiosensitive substrate.

Robustness of Deep Learning Algorithm to Varying Imaging Conditions in Detecting Low Contrast Objects In Computed Tomography Phantom Images: in Comparison to 12 Radiologists.

The performance of deep learning algorithm (DLA) to that of radiologists was compared in detecting low contrast objects in CT phantom images under various imaging conditions. For training, 10,000 images were created using American College of Radiology CT phantom as the background. In half of the images, objects of 3–20 mm size and 5–30 HU contrast difference were generated in random locations. Binary responses were used as the ground truth. For testing, 640 images of Catphan® phantom were used, half of which had objects of either 5 or 9 mm size with 10 HU contrast difference. Twelve radiologists evaluated the presence of objects on a five-point scale. The performances of the DLA and radiologists were compared across different imaging conditions in terms of area under receiver operating characteristics curve (AUC). Multi-reader multi-case AUC and Hanley and McNeil tests were used. We performed post-hoc analysis using bootstrapping and verified that the DLA is less affected by the changing imaging conditions. The AUC of DLA was consistently higher than those of the radiologists across different imaging conditions (p < 0.0001), and it was less affected by varying imaging conditions. The DLA outperformed the radiologists and showed more robust performance under varying imaging conditions.

Image denoising by transfer learning of generative adversarial network for dental CT

The successful development of the image denoising techniques for low-dose computed tomography (LDCT) was largely owing to the public-domain availability of spatially-aligned high- and low-dose CT image pairs. Even though low-dose CT scans are also highly desired in dental imaging, public-domain databases of dental CT image pairs have not been established yet. In this paper, we propose a dental CT image denoising method based on the transfer learning of a generative adversarial network (GAN) from the public-domain CT images. We trained a generative adversarial network with the Wasserstein loss function (WGAN) using 5,100 high- and low-dose medical CT image pairs of human chest and abdomen. For the generative network of GAN, we used the U-net structure of five stages to exploit its high computational efficiency. After training the proposed network, named U-WGAN, we fine-tuned the network with 3,006 dental CT image pairs of two different human skull phantoms. For the high- and low-dose scans of the phantoms, we set the tube current of the dental CT to 10 mA and 4 mA, respectively, with the tube voltage set to 90 kVp in both scans. We applied the trained network to denoising of low-dose dental CT images of dental phantoms and adult humans. The U-net processed images showed over-smoothing effects even though U-net had a good performance in the quantitative metrics. U-WGAN showed similar denoising performance to WGAN, but it reduced the computation time of WGAN by a factor of 10. The fine-tuning procedure in the transfer learning scheme enhanced the network performance in terms of the quantitative metrics, and it also improved visual appearance of the processed images. Even though the number of fine-tuning images was very limited in this study, we think the transfer learning scheme can be a good option for developing deep learning networks for dental CT image denoising.

Evaluating medical images using deep convolutional neural networks: A simulated CT phantom image study.

Applied research on artificial intelligence, mainly in deep learning, is widely performed. If medical images can be evaluated using artificial intelligence, this could substantially improve examination efficiency. We investigated an evaluation system for medical images with different noise characteristics using a deep convolutional neural network. Simulated computed tomography images are the targets of the system. We used an AlexNet trained with natural images for the deep convolutional neural network and a support vector machine for classification. Synthetic computed tomography images with circular and rectangular signal bodies at different levels of contrast and added Gaussian noise were used for training and testing. Two transfer learning methods were tested: classification by a re-trained support vector machine using the AlexNet features, and a method that fine-tuned the deep convolutional neural network. Using the first method, all the test image noise levels could be classified correctly. The fine-tuning method achieved an accuracy rate of 92.6%. An image quality evaluation method using artificial intelligence will be useful for clinical images and different image quality indices in the future.