The influence of image reconstruction methods on the diagnosis of pulmonary emphysema with convolutional neural network.

This study investigated the influence of iterative reconstruction (IR) methods on computed tomography (CT) images when training convolutional neural network (CNN) models to diagnose pulmonary emphysema. To evaluate the influence of the IR algorithm on CNN, the present study comprised two steps: the comparison of noise reduction by IR algorithms using phantom examinations and the change in performance of CNN with IR algorithms using patient data. We retrospectively analyzed 97 patients. Raw CT data were reconstructed using the filtered back-projection (FBP) and adaptive statistical iterative reconstruction V (ASIR-V) algorithms with blending levels of 30%, 50%, and 70%. The models were trained using reconstructed CT images and were named the FBP, ASIR-V30, ASIR-V50, and ASIR-V70 models. The mean and the standard deviation of the CT values were 11.3 ± 21.2 at FBP, 11.0 ± 17.3 at ASIR-V30, 11.0 ± 14.4 at ASIR-V50, and 11.0 ± 11.8 at ASIR-V70. For all the evaluation metrics, the best values were obtained with the FBP model applied to the ASIR-V70 test images. The worst values were obtained with the ASIR-V70 model applied to the FBP test images. The model trained with FBP images exhibited significantly better performance than the models trained using IR images. The reduction in image noise with the IR algorithm on the test images contributed to improving the accuracy of the classification of emphysema subtypes using CNN.

Virtual clinical trial based on outcome modeling with iteratively redistributed extrapolation data.

Virtual clinical trials (VCTs) can potentially simulate clinical trials on a computer, but their application with a limited number of past clinical cases is challenging due to the biased estimation of the statistical population. In this study, we developed ExMixup, a novel training technique based on machine learning, using iteratively redistributed extrapolated data. Information obtained from 100 patients with prostate cancer and 385 patients with oropharyngeal cancer was used to predict the recurrence after radiotherapy. Model performance was evaluated by developing outcome prediction models based on three types of training methods: training with original data (baseline), interpolation data (Mixup), and interpolation + extrapolation data (ExMixup). Two types of VCTs were conducted to predict the treatment response of patients with distinct characteristics compared to the training data obtained from patient cohorts categorized under risk classification or cancer stage. The prediction models developed with ExMixup yielded concordance indices (95% confidence intervals) of 0.751 (0.719-0.818) and 0.752 (0.734-0.785) for VCTs on the prostate and oropharyngeal cancer datasets, respectively, which significantly outperformed the baseline and Mixup models (P < 0.01). The proposed approach could enhance the ability of VCTs to predict treatment results in patients excluded from past clinical trials.

Evaluation of a Monte Carlo-based algorithm for the influence of totally implantable venous access ports in external radiation therapy.

PURPOSE: This study aimed to assess whether a Monte Carlo (MC)-based algorithm reflects the influence of totally implantable venous access ports (TIVAPs) in external radiation therapy. MATERIALS AND METHODS: The present study comprised two steps: experimental measurements of depth doses and surface doses with and without TIVAPs and calculation with an MC-based algorithm. RESULTS: The TIVAP-associated maximum dose reduction compared with the dose at the same depths without TIVAPs was 7.8% at 4 MV, 6.9% at 6 MV, and 5.7% at 10 MV in measurement, and 7.4% at 4 MV, 6.6% at 6 MV, and 5.5% at 10 MV in calculation. Relative surface doses were higher with TIVAPs made of titanium, due to a higher fluence of backscattered electrons from the TIVAPs, than with plastic TIVAPs. There were no significant differences in the relative differences between the measured and calculated doses of the titanium TIVAP group and the plastic TIVAP group at 4 MV (p = 0.99), 6 MV (p = 0.67), and 10 MV (p = 0.54). CONCLUSION: TIVAPs caused target dose reductions and dose increase near the TIVAP, especially when made of titanium. The influences are reflected in the MC-based algorithm.

Fully automated dose prediction using generative adversarial networks in prostate cancer patients.

PURPOSE: Although dose prediction for intensity modulated radiation therapy (IMRT) has been accomplished by a deep learning approach, delineation of some structures is needed for the prediction. We sought to develop a fully automated dose-generation framework for IMRT of prostate cancer by entering the patient CT datasets without the contour information into a generative adversarial network (GAN) and to compare its prediction performance to a conventional prediction model trained from patient contours. METHODS: We propose a synthetic approach to translate patient CT datasets into a dose distribution for IMRT. The framework requires only paired-images, i.e., patient CT images and corresponding RT-doses. The model was trained from 81 IMRT plans of prostate cancer patients, and then produced the dose distribution for 9 test cases. To compare its prediction performance to that of another trained model, we created a model trained from structure images. Dosimetric parameters for the planning target volume (PTV) and organs at risk (OARs) were calculated from the generated and original dose distributions, and mean differences of dosimetric parameters were compared between the CT-based model and the structure-based model. RESULTS: The mean differences of all dosimetric parameters except for D98% and D95% for PTV were within approximately 2% and 3% of the prescription dose for OARs in the CT-based model, while the differences in the structure-based model were within approximately 1% for PTV and approximately 2% for OARs, with a mean prediction time of 5 seconds per patient. CONCLUSIONS: Accurate and rapid dose prediction was achieved by the learning of patient CT datasets by a GAN-based framework. The CT-based dose prediction could reduce the time required for both the iterative optimization process and the structure contouring, allowing physicians and dosimetrists to focus their expertise on more challenging cases.

Fast Statistical Iterative Reconstruction for Mega-voltage Computed Tomography.

Statistical iterative reconstruction is expected to improve the image quality of computed tomography (CT). However, one of the challenges of iterative reconstruction is its large computational cost. The purpose of this review is to summarize a fast iterative reconstruction algorithm by optimizing reconstruction parameters. Megavolt projection data was acquired from a TomoTherapy system and reconstructed using in-house statistical iterative reconstruction algorithm. Total variation was used as the regularization term and the weight of the regularization term was determined by evaluating signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and visual assessment of spatial resolution using Gammex and Cheese phantoms. Gradient decent with an adaptive convergence parameter, ordered subset expectation maximization (OSEM), and CPU/GPU parallelization were applied in order to accelerate the present reconstruction algorithm. The SNR and CNR of the iterative reconstruction were several times better than that of filtered back projection (FBP). The GPU parallelization code combined with the OSEM algorithm reconstructed an image several hundred times faster than a CPU calculation. With 500 iterations, which provided good convergence, our method produced a 512 × 512 pixel image within a few seconds. The image quality of the present algorithm was much better than that of FBP for patient data. J. Med. Invest. 67 : 30-39, February, 2020.

A convolution neural network for higher resolution dose prediction in prostate volumetric modulated arc therapy.

PURPOSE: This study aims to investigate the feasibility of using convolutional neural networks to predict an accurate and high resolution dose distribution from an approximated and low resolution input dose. METHODS: Sixty-six patients were treated for prostate cancer with VMAT. We created the treatment plans using the Acuros XB algorithm with 2 mm grid size, followed by the dose calculated using the anisotropic analytical algorithm with 5 mm grid with the same plan parameters. U-net model was used to predict 2 mm grid dose from 5 mm grid dose. We investigated the two models differing for the training data used as input, one used just the low resolution dose (D model) and the other combined the low resolution dose with CT data (DC model). Dice similarity coefficient (DSC) was calculated to ascertain how well the shape of the dose-volume is matched. We conducted gamma analysis for the following: DVH from the two models and the reference DVH for all prostate structures. RESULTS: The DSC values in the DC model were significantly higher than those in the D model (p < 0.01). For the CTV, PTV, and bladder, the gamma passing rates in the DC model were significantly higher than those in the D model (p < 0.002-0.02). The mean doses in the CTV and PTV for the DC model were significantly better matched to those in the reference dose (p < 0.0001). CONCLUSIONS: The proposed U-net model with dose and CT image used as input predicted more accurate dose.

Optimization of treatment strategy by using a machine learning model to predict survival time of patients with malignant glioma after radiotherapy.

The purpose of this study was to predict the survival time of patients with malignant glioma after radiotherapy with high accuracy by considering additional clinical factors and optimize the prescription dose and treatment duration for individual patient by using a machine learning model. A total of 35 patients with malignant glioma were included in this study. The candidate features included 12 clinical features and 192 dose-volume histogram (DVH) features. The appropriate input features and parameters of the support vector machine (SVM) were selected using the genetic algorithm based on Akaike's information criterion, i.e. clinical, DVH, and both clinical and DVH features. The prediction accuracy of the SVM models was evaluated through a leave-one-out cross-validation test with residual error, which was defined as the absolute difference between the actual and predicted survival times after radiotherapy. Moreover, the influences of various values of prescription dose and treatment duration on the predicted survival time were evaluated. The prediction accuracy was significantly improved with the combined use of clinical and DVH features compared with the separate use of both features (P < 0.01, Wilcoxon signed rank test). Mean ± standard deviation of the leave-one-out cross-validation using the combined clinical and DVH features, only clinical features and only DVH features were 104.7 ± 96.5, 144.2 ± 126.1 and 204.5 ± 186.0 days, respectively. The prediction accuracy could be improved with the combination of clinical and DVH features, and our results show the potential to optimize the treatment strategy for individual patients based on a machine learning model.

Evaluation of organ-at-risk dose reduction with jaw tracking technique in flattening filter-free beams in lung stereotactic body radiation therapy.

PURPOSE: (i) to investigate the capability of organ-at-risk (OAR) dose reduction with the jaw tracking (JT) technique in flattening filter-free (FFF) beams in lung stereotactic body radiation therapy (SBRT), (ii) to propose a novel metric to quantify the jaw movements during JT, and (iii) to examine the relationships between the quantified jaw movements and reduction rate of OAR doses. METHODS: The individual SBRT plans with volumetric modulated arc therapy using the JT technique (JT-VMAT) and VMAT plans with a fixed jaw (FJ-VMAT) were created for 15 patients, and dosimetric parameters were compared. A jaw tracking complexity score (JTCS) was defined and compared with the multi-leaf collimator (MLC) modulation complexity score (MCS). The correlations between the JTCS and reduction rate of OAR doses were examined. RESULTS: The decrease of OARs doses was statistically significant in the JT-VMAT plans (1.2% in V20 of the lung and <1% in all other OARs). The correlations between the JTCS and MCS were not significant. There were significant correlations between the JTCS and the reduction rates in V20, V2.5, and Dmean of the lung, D1% of the spinal cord, and D90% of the body. CONCLUSIONS: A significant decrease of dosimetric parameters of OARs was found with JT-VMAT in FFF beams. This reduction is very small and probably not clinically relevant. JTCS, a novel metric to quantify the jaw movements during JT, was proposed, and the complexity of jaw movements did not correlate with that of the movements of MLC leaves. There were significant correlations between the JTCS and some dosimetric parameters of OARs.

Deep convolutional neural network for reduction of contrast-enhanced region on CT images.

This study aims to produce non-contrast computed tomography (CT) images using a deep convolutional neural network (CNN) for imaging. Twenty-nine patients were selected. CT images were acquired without and with a contrast enhancement medium. The transverse images were divided into 64 × 64 pixels. This resulted in 14 723 patches in total for both non-contrast and contrast-enhanced CT image pairs. The proposed CNN model comprises five two-dimensional (2D) convolution layers with one shortcut path. For comparison, the U-net model, which comprises five 2D convolution layers interleaved with pooling and unpooling layers, was used. Training was performed in 24 patients and, for testing of trained models, another 5 patients were used. For quantitative evaluation, 50 regions of interest (ROIs) were selected on the reference contrast-enhanced image of the test data, and the mean pixel value of the ROIs was calculated. The mean pixel values of the ROIs at the same location on the reference non-contrast image and the predicted non-contrast image were calculated and those values were compared. Regarding the quantitative analysis, the difference in mean pixel value between the reference contrast-enhanced image and the predicted non-contrast image was significant (P < 0.0001) for both models. Significant differences in pixels (P < 0.0001) were found using the U-net model; in contrast, there was no significant difference using the proposed CNN model when comparing the reference non-contrast images and the predicted non-contrast images. Using the proposed CNN model, the contrast-enhanced region was satisfactorily reduced.

Cone-beam CT reconstruction for non-periodic organ motion using time-ordered chain graph model.

PURPOSE: The purpose of this study is to introduce the new concept of a four-dimensional (4D) cone-beam computed tomography (CBCT) reconstruction approach for non-periodic organ motion in cooperation with the time-ordered chain graph model (TCGM) and to compare it with previously developed methods such as total variation-based compressed sensing (TVCS) and prior-image constrained compressed sensing (PICCS). MATERIALS AND METHODS: Our proposed reconstruction is based on a model including the constraint originating from the images of neighboring time phases. Namely, the reconstructed time-series images depend on each other in this TCGM scheme, and the time-ordered images are concurrently reconstructed in the iterative reconstruction approach. In this study, iterative reconstruction with the TCGM was carried out with 90° projection ranges. The images reconstructed by the TCGM were compared with the images reconstructed by TVCS (200° projection ranges) and PICCS (90° projection ranges). Two kinds of projection data sets-an elliptic-cylindrical digital phantom and two clinical patients' data-were used. For the digital phantom, an air sphere was contained and virtually moved along the longitudinal axis by 3 cm/30 s and 3 cm/60 s; the temporal resolution was evaluated by measuring the penumbral width of the air sphere. The clinical feasibility of the non-periodic time-ordered 4D CBCT image reconstruction was examined with the patient data in the pelvic region. RESULTS: In the evaluation of the digital-phantom reconstruction, the penumbral widths of the TCGM yielded the narrowest result; the results obtained by PICCS and TCGM using 90° projection ranges were 2.8% and 18.2% for 3 cm/30 s, and 5.0% and 23.1% for 3 cm/60 s narrower than that of TVCS using 200° projection ranges. This suggests that the TCGM has a better temporal resolution, whereas PICCS seems similar to TVCS. These reconstruction methods were also compared using patients' projection data sets. Although all three reconstruction results showed motion related to rectal gas or stool, the result obtained by the TCGM was visibly clearer with less blurring. CONCLUSION: The TCGM is a feasible approach to visualize non-periodic organ motion. The digital-phantom results demonstrated that the proposed method provides 4D image series with a better temporal resolution compared to TVCS and PICCS. The clinical patients' results also showed that the present method enables us to visualize motion related to rectal gas and flatus in the rectum.

Whole-Body Distribution of Leukemia and Functional Total Marrow Irradiation Based on FLT-PET and Dual-Energy CT.

This report describes a multimodal whole-body 3'-deoxy-3'[(18)F]-fluorothymidine positron emission tomography (FLT-PET) and dual-energy computed tomography (DECT) method to identify leukemia distribution within the bone marrow environment (BME) and to develop disease- and/or BME-specific radiation strategies. A control participant and a newly diagnosed patient with acute myeloid leukemia prior to induction chemotherapy were scanned with FLT-PET and DECT. The red marrow (RM) and yellow marrow (YM) of the BME were segmented from DECT using a basis material decomposition method. Functional total marrow irradiation (fTMI) treatment planning simulations were performed combining FLT-PET and DECT imaging to differentially target irradiation to the leukemia niche and the rest of the skeleton. Leukemia colonized both RM and YM regions, adheres to the cortical bone in the spine, and has enhanced activity in the proximal/distal femur, suggesting a potential association of leukemia with the BME. The planning target volume was reduced significantly in fTMI compared with conventional TMI. The dose to active disease (standardized uptake value >4) was increased by 2-fold, while maintaining doses to critical organs similar to those in conventional TMI. In conclusion, a hybrid system of functional-anatomical-physiological imaging can identify the spatial distribution of leukemia and will be useful to both help understand the leukemia niche and develop targeted radiation strategies.

Use of dual-energy computed tomography to measure skeletal-wide marrow composition and cancellous bone mineral density.

Temporal and spatial variations in bone marrow adipose tissue (MAT) can be indicative of several pathologies and confound current methods of assessing immediate changes in bone mineral remodeling. We present a novel dual-energy computed tomography (DECT) method to monitor MAT and marrow-corrected volumetric BMD (mcvBMD) throughout the body. Twenty-three cancellous skeletal sites in 20 adult female cadavers aged 40-80 years old were measured using DECT (80 and 140 kVp). vBMD was simultaneous recorded using QCT. MAT was further sampled using MRI. Thirteen lumbar vertebrae were then excised from the MRI-imaged donors and examined by microCT. After MAT correction throughout the skeleton, significant differences (p < 0.05) were found between QCT-derived vBMD and DECT-derived mcvBMD results. McvBMD was highly heterogeneous with a maximum at the posterior skull and minimum in the proximal humerus (574 and 0.7 mg/cc, respectively). BV/TV and BMC have a nearly significant correlation with mcvBMD (r = 0.545, p = 0.057 and r = 0.539, p = 0.061, respectively). MAT assessed by DECT showed a significant correlation with MRI MAT results (r = 0.881, p < 0.0001). Both DECT- and MRI-derived MAT had a significant influence on uncorrected vBMD (r = -0.86 and r = -0.818, p ≤ 0.0001, respectively). Conversely, mcvBMD had no correlation with DECT- or MRI-derived MAT (r = 0.261 and r = 0.067). DECT can be used to assess MAT while simultaneously collecting mcvBMD values at each skeletal site. MAT is heterogeneous throughout the skeleton, highly variable, and should be accounted for in longitudinal mcvBMD studies. McvBMD accurately reflects the calcified tissue in cancellous bone.

Evaluation of Functional Marrow Irradiation Based on Skeletal Marrow Composition Obtained Using Dual-Energy Computed Tomography.

PURPOSE: To develop an imaging method to characterize and map marrow composition in the entire skeletal system, and to simulate differential targeted marrow irradiation based on marrow composition. METHODS AND MATERIALS: Whole-body dual energy computed tomography (DECT) images of cadavers and leukemia patients were acquired, segmented to separate bone marrow components, namely, bone, red marrow (RM), and yellow marrow (YM). DECT-derived marrow fat fraction was validated using histology of lumbar vertebrae obtained from cadavers. The fractions of RM (RMF = RM/total marrow) and YMF were calculated in each skeletal region to assess the correlation of marrow composition with sites and ages. Treatment planning was simulated to target irradiation differentially at a higher dose (18 Gy) to either RM or YM and a lower dose (12 Gy) to the rest of the skeleton. RESULTS: A significant correlation between fat fractions obtained from DECT and cadaver histology samples was observed (r=0.861, P<.0001, Pearson). The RMF decreased in the head, neck, and chest was significantly inversely correlated with age but did not show any significant age-related changes in the abdomen and pelvis regions. Conformity of radiation to targets (RM, YM) was significantly dependent on skeletal sites. The radiation exposure was significantly reduced (P<.05, t test) to organs at risk (OARs) in RM and YM irradiation compared with standard total marrow irradiation (TMI). CONCLUSIONS: Whole-body DECT offers a new imaging technique to visualize and measure skeletal-wide marrow composition. The DECT-based treatment planning offers volumetric and site-specific precise radiation dosimetry of RM and YM, which varies with aging. Our proposed method could be used as a functional compartment of TMI for further targeted radiation to specific bone marrow environment, dose escalation, reduction of doses to OARs, or a combination of these factors.

Fast Megavoltage Computed Tomography: A Rapid Imaging Method for Total Body or Marrow Irradiation in Helical Tomotherapy.

PURPOSE: Megavoltage computed tomographic (MVCT) imaging has been widely used for the 3-dimensional (3-D) setup of patients treated with helical tomotherapy (HT). One drawback of MVCT is its very long imaging time, the result of slow couch speeds of approximately 1 mm/s, which can be difficult for the patient to tolerate. We sought to develop an MVCT imaging method allowing faster couch speeds and to assess its accuracy for image guidance for HT. METHODS AND MATERIALS: Three cadavers were scanned 4 times with couch speeds of 1, 2, 3, and 4 mm/s. The resulting MVCT images were reconstructed using an iterative reconstruction (IR) algorithm with a penalty term of total variation and with a conventional filtered back projection (FBP) algorithm. The MVCT images were registered with kilovoltage CT images, and the registration errors from the 2 reconstruction algorithms were compared. This fast MVCT imaging was tested in 3 cases of total marrow irradiation as a clinical trial. RESULTS: The 3-D registration errors of the MVCT images reconstructed with the IR algorithm were smaller than the errors of images reconstructed with the FBP algorithm at fast couch speeds (2, 3, 4 mm/s). The scan time and imaging dose at a speed of 4 mm/s were reduced to 30% of those from a conventional coarse mode scan. For the patient imaging, faster MVCT (3 mm/s couch speed) scanning reduced the imaging time and still generated images useful for anatomic registration. CONCLUSIONS: Fast MVCT with the IR algorithm is clinically feasible for large 3-D target localization, which may reduce the overall time for the treatment procedure. This technique may also be useful for calculating daily dose distributions or organ motion analyses in HT treatment over a wide area. Automated integration of this imaging is at least needed to further assess its clinical benefits.

Characterization of deformation and physical force in uniform low contrast anatomy and its impact on accuracy of deformable image registration.

PURPOSE: Little is known about the effect of force on organ deformation and consequently its impact on precision dose delivery. The purpose of this study was to evaluate the fundamental relationship between anatomic deformation and its causative physical force to ascertain if a threshold limit exists for deformable image registration (DIR) accuracy in uniform low contrast anatomy, beyond which its applicability may be clinically inappropriate. METHODS: To simulate a simplified model, a tissue equivalent deformable bladder phantom with 21 implanted fiducial markers was developed using a viscoelastic polymer. The bladder phantom was deformed by applying a force in increments from 10 to 70 N. DIR accuracy was studied using intensity based mim and Velocity B-spline algorithms by comparing the 3D vector of the 21 marker locations at the original target image with the synthetically derived marker positions from each target image obtained from DIR. RESULTS: The relationship between applied force in 1D deformation along the axis of applied force and 3D deformation of the phantom showed a linear response. The maximum and average displacements of markers exhibited a nonlinear response to the applied force. In the absence of implanted markers, DIR performance was suboptimal with a threshold limit of only 20 N (5 mm deformation) beyond which the average marker error was ≥3 mm. DIR performance improved significantly with the addition of only one marker for the intensity based mim algorithm. In contrast, the Velocity B-spline algorithm showed reduced sensitivity to the number of markers introduced in both the source and target images. CONCLUSIONS: The limits of applicability of DIR are strongly dependent on the magnitude of deformation. There is a threshold limit beyond which the accuracy of DIR fails in uniform low contrast anatomy. The sensitivity of the DIR performance to the number of fiducial markers present indicates that if DIR performance is solely assessed with the contrast rich features present in clinical anatomy, the results may not be reflective of the true DIR performance in uniform low contrast anatomy.