Browsing by Subject "Deformable image registration"
- Results Per Page
- Sort Options
Item Open Access Clinical and Research Applications of 3D Dosimetry(2015-01-01) Juang, TitaniaQuality assurance (QA) is a critical component of radiation oncology medical physics for both effective treatment and patient safety, particularly as innovations in technology allow movement toward advanced treatment techniques that require increasingly higher accuracy in delivery. Comprehensive 3D dosimetry with PRESAGE® 3D dosimeters read out via optical CT has the potential to detect errors that would be missed by current systems of measurement, and thereby improve the rigor of current QA techniques through providing high-resolution, full 3D verification for a wide range of clinical applications. The broad objective of this dissertation research is to advance and strengthen the standards of QA for radiation therapy, both by driving the development and optimization of PRESAGE® 3D dosimeters for specific clinical and research applications and by applying the technique of high resolution 3D dosimetry toward addressing clinical needs in the current practice of radiation therapy. The specific applications that this dissertation focuses on address several topical concerns: (1) increasing the quality, consistency, and rigor of radiation therapy delivery through comprehensive 3D verification in remote credentialing evaluations, (2) investigating a reusable 3D dosimeter that could potentially facilitate wider implementation of 3D dosimetry through improving cost-effectiveness, and (3) validating deformable image registration (DIR) algorithms prior to clinical implementation in dose deformation and accumulation calculations.
3D Remote Dosimetry: The feasibility of remote high-resolution 3D dosimetry with the PRESAGE®/Optical-CT system was investigated using two nominally identical optical-CT scanners for 3D dosimetry were constructed and placed at the base (Duke University) and remote (IROC Houston) institutions. Two formulations of PRESAGE® (SS1, SS2) were investigated with four unirradiated PRESAGE® dosimeters imaged at the base institution, then shipped to the remote institution for planning and irradiation. After each dosimeter was irradiated with the same treatment plan and subsequently read out by optical CT at the remote institution, the dosimeters were shipped back to the base institution for remote dosimetry readout 3 days post-irradiation. Measured on-site and remote relative 3D dose distributions were registered to the Pinnacle dose calculation, which served as the reference distribution for 3D gamma calculations with passing criteria of 5%/2mm, 3%/3mm, and 3%/2mm with a 10% dose threshold. Gamma passing rates, dose profiles, and dose maps were used to assess and compare the performance of both PRESAGE® formulations for remote dosimetry. Both PRESAGE® formulations under study maintained high linearity of dose response (R2>0.996) over 14 days with response slope consistency within 4.9% (SS1) and 6.6% (SS2). Better agreements between the Pinnacle plan and dosimeter readout were observed in PRESAGE® formulation SS2, which had higher passing rates and consistency between immediate and remote results at all metrics. This formulation also demonstrated a relative dose distribution that remained stable over time. These results provide a foundation for future investigations using remote dosimetry to study the accuracy of advanced radiation treatments.
A Reusable 3D Dosimeter: New Presage-RU formulations made using a lower durometer polyurethane matrix (Shore hardness 30-50A) exhibit a response that optically clears following irradiation and opens up the potential for reirradiation and dosimeter reusability. This would have the practical benefit of improving cost-effectiveness and thereby facilitating the wider implementation of comprehensive, high resolution 3D dosimetry. Three formulations (RU-3050-1.7, RU-3050-1.5, and RU-50-1.5) were assessed with multiple irradiations of both small volume samples and larger volume dosimeters, then characterized and evaluated for dose response sensitivity, optical clearing, dose-rate independence, dosimetric accuracy, and the effects of reirradiation on dose measurement. The primary shortcoming of these dosimeters was the discovery of age-dependent gradients in dose response sensitivity, which varied dose response by as much as 30% and prevented accurate measurement. This is unprecedented in the standard formulations and presumably caused by diffusion of a desensitizing agent into the lower durometer polyurethane. The effect of prior irradiation on the dosimeters would also be a concern as it was seen that the relative amount of dose delivered to any given region of the dosimeter will affect subsequent sensitivity in that area, which would in effect create spatially-dependent variable dose sensitivities throughout the dosimeter based on the distributions of prior irradiations. While a successful reusable dosimeter may not have been realized from this work, these studies nonetheless contributed useful information that will affect future development, including in the area of deformable dosimetry, and provide a framework for future reusable dosimeter testing.
Validating Deformable Image Registration Algorithms: Deformable image registration (DIR) algorithms are used for multi-fraction dose accumulation and treatment response assessment for adaptive radiation therapy, but the accuracy of these methods must be investigated prior to clinical implementation. 12 novel deformable PRESAGE® 3D dosimeter formulations were introduced and characterized for potential use in validating DIR algorithms by providing accurate, ground-truth deformed dose measurement for comparison to DIR-predicted deformed dose distributions. Two commercial clinical DIR software algorithms were evaluated for dose deformation accuracy by comparison against a measured deformed dosimeter dose distribution. This measured distribution was obtained by irradiating a dosimeter under lateral compression, then releasing it from compression so that it could return to its original geometry. The dose distribution within the dosimeter deformed along with the dosimeter volume as it regained to its original shape, thus providing a measurable ground truth deformed dose distribution. Results showed that intensity-based DIR algorithms produce high levels of error and physically unrealistic deformations when deforming a homogeneous structure; this is expected as lack of internal structure is challenging for intensity-based DIR algorithms to deform accurately as they rely on matching fairly closely spaced heterogeneous intensity features. A biomechanical, intensity-independent DIR algorithm demonstrated substantially closer agreement to the measured deformed dose distribution with 3D gamma passing rates (3%/3mm) in the range of 90-91%. These results underscore the necessity and importance of validating DIR algorithms for specific clinical scenarios prior to clinical implementation.
Item Open Access Development of Deep Learning Models for Deformable Image Registration (DIR) in the Head and Neck Region(2020) Amini, AlaDeformable image registration (DIR) is the process of registering two or more images to a reference image by minimizing local differences across the entire image. DIR is conventionally performed using iterative optimization-based methods, which are time-consuming and require manual parameter tuning. Recent studies have shown that deep learning methods, most importantly convolutional neural networks (CNNs), can be employed to address the DIR problem. In this study, we propose two deep learning frameworks to perform the DIR task in an unsupervised approach for CT-to-CT deformable registration of the head and neck region. Given that head and neck cancer patients might undergo severe weight loss over the course of their radiation therapy treatment, DIR in this region becomes an important task. The first proposed deep learning framework contains two scales, where both scales are based on freeform deformation, and are trained based on minimizing a dissimilarity intensity-based metrics, while encouraging the deformed vector field (DVF) smoothness. The two scales were first trained separately in a sequential manner, and then combined in a two-scale joint training framework for further optimization. We then developed a transfer learning technique to improve the DIR accuracy of the proposed deep learning networks by fine-tuning a pre-trained group-based model into a patient-specific model to optimize its performance for individual patients. We showed that by utilizing as few as two prior CT scans of a patient, the performance of the pretrained model described above can be improved yielding more accurate DIR results for individual patients. The second proposed deep learning framework, which also consists of two scales, is a hybrid DIR method using B-spline deformation modeling and deep learning. In the first scale, deformation of control points are learned by deep learning and initial DVF is estimated using B-spline interpolation to ensure smoothness of the initial estimation. Second scale model of the second framework is the same as that in the first framework. In our study, the networks were trained and evaluated using public TCIA HNSCC-3DCT for the head and neck region. We showed that our DIR results of our proposed networks are comparable to conventional DIR methods while being several orders of magnitude faster (about 2 to 3 seconds), making it highly applicable for clinical applications.
Item Embargo Development of Imaging-Based Models for Analyzing the Spatiotemporal Function of Intervertebral Discs(2023) Coppock, James AveryLumbar intervertebral discs (IVD) play a critical role in facilitating the mobility and load-bearing functionality of the spine. Consequently, degeneration of the IVDs has been linked to the development of low back pain (LBP), a leading cause of disability in the world. While the pathomechanisms leading to the development of IVD degeneration and LBP are heterogeneous and often difficult to discern, it is believed that the changes in IVD function (i.e., mechanics, composition, tissue structure) may be closely related to the development of discogenic LBP. Specifically, because the IVD has a limited capacity to repair itself, disruptions to IVD tissue structures and biochemical composition may enable nervous tissue innervation into the IVD, potentiating the development of discogenic LBP. However, because our ability to study these changes in vivo remains limited, it remains unknown whether or not we can leverage the study of IVD function to identify risk factors associated with the development of LBP prior to their transition to a painful state. Accordingly, the overarching goal of this work is to develop non-invasive imaging techniques which may be used to perform spatiotemporal analyses of IVD kinematics and composition in vivo. Building upon prior work in our lab, Specific Aim 1 of this proposed research first seeks to develop a controlled methods to investigate the links between IVD function, composition and LBP by examining the in vivo response of IVDs to controlled dynamic loading in asymptomatic individuals. Using data generated in the prior aim, Specific Aim 2 then seeks to first develop and validate an image-segmentation method which enables precise kinematic analysis of the IVD to be carried out in an automated fashion, in vivo. Subsequently, Specific Aim 2 then seeks expand our current ability to characterize IVD function in response to dynamic activity by developing and validating a novel methodology for evaluating three-dimensional (3D) internal spatiotemporal changes in IVD kinematics using a novel deep-learning-based deformable image registration network. This dissertation is organized as a collection of original research articles which were conducted during my time as a PhD student in the Musculoskeletal Bioengineering Laboratory. The first of these (Chapter 3 - Increasing BMI Increases Lumbar Intervertebral Disc Deformation Following A Treadmill Walking Stress Test) was published in the Journal of Biomechanics (Coppock et al., 2021) in May 2021. The second of these (Chapter 4 - In vivo Intervertebral Disc Mechanical Deformation Following a Treadmill Walking “Stress Test” is Inversely Related to T1rho Relaxation Time) was published in the Osteoarthritis and Cartilage (Coppock et. al, 2022). The third, and fourth manuscripts are currently under review (Chapter 5 - Automated Segmentation and Prediction of Intervertebral Disc Morphology and Uniaxial Deformations from MRI; Chapter 6 - In Vivo Analysis of Intervertebral Disc Mechanics Using a Diffeomorphic Deep-Learning Approach. Chapter 7 - The Effects of a 6-month Weight Loss Intervention on Physical Function and Serum Biomarkers in Older Adults with and without Osteoarthritis - is published in Osteoarthritis and Cartilage, Open.
Item Open Access Evaluating Dose Summation in Gynecological Brachytherapy(2015) Poplawski, Linda JeanPurpose Gynecological malignancies present challenges in determining an appropri- ate volumetric dose due to the highly variable physiologic activity of the surrounding tissue. Because of the high doses used in brachytherapy, surrounding structures have the potential to move around in the dose region and receive unknown amounts of radiation. Deformable image registration could overcome challenges in determining the true delivered dose through a dose accumulation process. This study uses two dose summation techniques to determine the efficacy of a deformation registration for the Syed applicators and cylinder applicators.
Methods and Materials Data for patients treated with a vaginal cylinder or Syed template were imported into the MIM software (Cleveland, OH). The bladder, rectum and applicator were contoured on each CT. The deformable registration was applied to structures only by masking other image data to a single intensity with the purpose of focusing the registration on the high dose area, as well as to minimize any uncertainty from the CT data. The deformable registration flow consisted of the following steps: 1) Using a different Hounsfield Unit, the CTs were masked so that each of the structures-of-interest (SOIs) had one unique intensity value; 2) Perform a rigid registration between two image sets with alignment based on the applicator position; 3) Perform a deformable registration; 4) Refine registration by using local manual alignment in area with large contour changes; 5) Repeat steps 1 to 3 to register the desired structure from all the subsequent fractions to the first fraction structure; 5) Transfer each deformed contoured to the first CT. The deformed structure accuracy was measured by a comparison to the first fraction using the dice similarity coefficient. Two dose summation techniques were investigated: a) deform the desired structure to one fraction and determine the fractional dose to these new structures, and b) deform the fractional dose to one CT and accumulate to find the total dose. Point doses, D2cc, were used as a comparison value for each method.
Results The cylinder set of patients had a DSC ranging from 0.82 to 0.96 for bladder and 0.82 to 0.94 for rectum. The contour deformation addition method has variations up to 23% from the initial clinical point dose for the cylinder applicators. The dose deformation accumulation method gave up to a 7.1% difference from the clinical point dose. The Syed applicator patient set has DSC ranging from 0.53 to 0.97 for the bladder and 0.75 and 0.95 for the rectum. These registrations' dose additions varied up to 34.78% and the dose deformation accumulation varied up to 34.97%.
Conclusions With the changing anatomy in brachytherapy, deforming the dose with the end point of dose summation leads to different volumetric doses then when dose is recalculated on deformed structures, raising concerns about the accuracy of the deformed dose. Dice Similarity Coefficients alone cannot be used to establish the accuracy of a deformation for brachy dose summation purpose.
Item Open Access Evaluation of Deformable Image Registration for Lung Motion Estimation using Hyperpolarized Gas Tagging MRI(2014) Huang, QijiePurpose: Hyperpolarized gas (HP) tagging MRI is a novel imaging technique for direct measurement of lung motion during breathing. This study aims to quantitatively evaluate the accuracy of deformable image registration (DIR) in lung motion estimation using HP tagging MRI as references.
Method: Three healthy subjects were imaged using the HP MR tagging, as well as a high-resolution 3D proton MR sequence (TrueFISP) at the end-of-inhalation (EOI) and the end-of-exhalation (EOE). Ground truth of lung motion and corresponding displacement vector field (tDVF) was derived from HP tagging MRI by manually tracking the displacement of tagging grids between EOI and EOE. Seven different DIR methods were applied to the high-resolution TrueFISP MR images (EOI and EOE) to generate the DIR-based DVFs (dDVF). The DIR methods include Velocity (VEL), MIM, Mirada, multi-grid B-spline from Elastix (MGB) and 3 other algorithms from DIRART toolbox (Double Force Demons (DFD), Improved Lucas-Kanade (ILK), and Iterative Optical Flow (IOF)). All registrations were performed by independent experts. Target registration error (TRE) was calculated as tDVF - dDVF. Analysis was performed for the entire lungs, and separately for the upper and lower lungs.
Results: Significant differences between tDVF and dDVF were observed. Besides the DFD and IOF algorithms, all other dDVFs showed similarity in deformation magnitude distribution but away from the ground truth. The average TRE for entire lung ranged 2.5-23.7mm (mean=8.8mm), depending on the DIR method and subject's breathing amplitude. Larger TRE (13.3-23.7mm) was found in subject with larger breathing amplitude of 45.6mm. TRE was greater in lower lung (2.5-33.9 mm, mean=12.4mm) than that in upper lung (2.5-11.9 mm, mean=5.8mm).
Conclusion: Significant differences were observed in lung motion estimation between the HP gas tagging MRI method and the DIR methods, especially when lung motion is large. Large variation among different DIR methods was also observed.
Item Open Access Evaluation of Lung Ventilation Maps with Hyperpolarized Gas Tagged Magnetic Resonance Imaging and its Application towards Respiratory Motion Modeling(2015) Cui, TaoranPurpose: Direct measurement of regional lung ventilation is often more favorable than the calculated ones using deformable image registration (DIR), which has been found susceptible to many uncertainties. This study aims to investigate the possible implementation of a novel imaging technique, 3D Hyperpolarized (HP) gas tagging MRI, to evaluate the regional ventilation of lung calculated with DIR-based algorithms, and to model the respiratory motion of lung towards 4D digital phantom application, respectively.
Method: Three healthy volunteers involved in this study underwent both 3D HP gas tagging MRI (t-MRI) and 3D proton true fast imaging with steady state precession (TrueFISP) MRI (p-MRI) during two consecutive breath-holds at the inhalation and the exhalation, respectively.
A ground truth of displacement vector field (DVF) was obtained by tracking the location of each tagged grid between the exhalation and the inhalation t-MRI images. Meanwhile, p-MRI images were registered with two commercial DIR algorithms, Velocity AI and MIM, to generate the corresponding DVFs. The regional ventilations were calculated as the Jacobian of each DVF. The cross-correlation and the mutual information were computed between either the DIR-based and the ground truth ventilations for comparison.
Regional ventilation of lung, as a physiological property, should remain invariable during the image acquisition. Therefore, in order to develop a realistic model for the respiratory motion of lung, the optimization of the DIR-based DVF was performed by imposing a regularization on the regional ventilation of lung. The robustness of the proposed model was evaluated by comparing the optimized DVF to the original DIR-based DVF in cross-correlation and mutual information.
Results: Compared to the ground truth result of the 3D HP gas tagging MRI, the regional ventilation maps calculated using the commercial DIR algorithms varied considerably for all the subjects of different respiration amplitudes. The ventilation derived from Velocity AI was preferable for the better spatial homogeneity and the larger accuracy, given by the higher average cross correlation (0.328 v.s. 0.262) and average mutual information (0.528 v.s. 0.323).
The optimized model represents a more realistic lung motion with the average deviation from the ground truth DVF reduced from 6.4 mm to 5.2 mm (p<0.05, student t-test) compared to the original DIR-based DVF. Furthermore, the accuracy of regional ventilation was also improved with larger cross correlation (0.98 v.s. 0.37) and mutual information values (5.84 v.s. 0.45).
Conclusion:
The results suggest that direct measurement of lung motion using 3D hyperpolarized gas tagging MRI may have the potential to assess, validate, and improve the DIR-based application. The proposed 4D motion model of lung also holds great promises for the implementation in a digital phantom, which can be applied in versatile researches for 4D radiation therapy of lung cancer.
Item Open Access Image-based longitudinal assessment of external beam radiation therapy for gynecological malignancies(2023) Eckrich, CarolynThis thesis consists of two studies. Study 1 is an assessment of dose-volume metrics of an 18F-FDG PET adaptive radiation therapy for vulvar and cervical cancer patients.Study 2 is an evaluation of cumulative dose distributions from external beam radiation therapy using CT-to-CBCT deformable image registration (DIR) for cervical cancer patients.
Study 1: Assessment of dose-volume metrics of an 18F-FDG PET adaptive radiation therapy for vulvar and cervical cancer patientsPurpose: Adaptive radiation therapy (ART) enables treatment to be modified with the goal of improving the dose distribution to the patient due to changes in anatomy. Fluorodeoxyglucose positron emission tomography and computed tomography (FDG-PET/CT) is used for staging, treatment planning, and assessing treatment response, but can also be used to adapt treatment. In an adaptive PET/CT study, an additional PET/CT scan is acquired for planning purposes after a certain prescribed dose has been delivered. The intratreatment PET/CT is used to re-contour the volumes and create a new treatment plan that is used to deliver the remaining dose for the treatment. The goal of adaptive radiation therapy (ART) is to reduce the dose to normal tissues while maintaining the prescribed dose to the adapted PTV. Materials and Methods: In this IRB-approved protocol, patients with vulvar and cervical cancer received a planning PET/CT and an intratreatment PET/CT. Radiation therapy consisted of either intensity modulated radiotherapy (IMRT) or volumetric modulated arc therapy (VMAT) with 1.8 Gy once daily to a total of 45 to 50.4 Gy and simultaneous integrated boosts (SIB) to involved pelvic or para-aortic (PA) lymph nodes. The primary tumor was treated to 64.4 to 66.4 Gy with sequential boosts for the vulvar cancer patients. Cervical cancer patients were boosted with brachytherapy. SIB dose ranged from a total of 64.4 Gy to 66.4 Gy in 25 fractions determined by the treating physician and organs-at risk (OAR) tolerance. An intratreatment PET/CT was obtained at 12-20 fractions when the delivered dose was between 30 to 36 Gy. All patients were re-planned with revised OAR, gross tumor volume (GTV) and planning target volume (PTV) contours. The same dose goals remained on the adapted plan. Dosimetric metrics for OARs were compared using the Wilcoxon signed rank test. The criteria for determining statistical significance was established as a p-value less than 0.05. Results: In the vulvar analysis, out of 20 eligible patients, ART resulted in significant reductions in OAR doses. For bladder, max dose (Dmax) median reduction (MR) was 1.1 Gy ((IQR 0.48 – 2.3 Gy), p < 0.001) and for D2cc MR was 1.5 Gy ((IQR 0.51 – 2.1 Gy), p < 0.001). For bowel, Dmax MR was 1.0 Gy ((IQR 0.11 – 2.9 Gy), p < 0.001), for D2cc MR was 0.39 Gy ((IQR 0.023 – 1.7 Gy), p < 0.001), and for D15cc MR was 0.19 Gy ((IQR 0.026 – 0.47 Gy), p = 0.002)). For rectum, mean dose (Dmean) MR was 0.66 Gy ((IQR 0.17 – 1.7 Gy) p = 0.006) and for D2cc MR was 0.46 Gy ((IQR 0.17 – 0.80 Gy), p = 0.006). Thirty-seven cervical patients were analyzed. ART resulted in significant reductions in OAR doses. For bladder, max dose (Dmax) median reduction (MR) was 0.89 Gy ((IQR 0.23 – 2.14 Gy), p = 0.001) and for D2cc MR was 0.38 Gy ((IQR 0.12 – 1.36 Gy), p<0.0001). For bowel, Dmax MR was 3.27 Gy ((IQR 0.50 – 5.41 Gy), p < 0.0001). For D2cc MR was 2.09 Gy ((IQR 0.30 – 4.97 Gy), p < 0.0001), and for D15cc MR was 0.57 Gy (IQR 0.22 – 2.07 Gy)). For rectum, Dmean MR was 0.13 Gy ((IQR 0.09 – 0.24 Gy) p = 0.0025), and for D2cc MR was 0.44 Gy ((IQR 0.14 – 1.02 Gy), p < 0.0001). Conclusions: Based on the analysis and response to ART of 20 eligible patients with vulvar cancer and 37 eligible patients with cervical cancer, it can be concluded that ART resulted in a significant reduction in OAR doses, including bladder, bowel, and rectum. Overall, these findings suggest that ART can effectively reduce the radiation dose to OARs and improve treatment outcomes for patients with gynecological cancers.
Study 2: Evaluation of cumulative dose distributions from external beam radiation therapy using CT-to-CBCT deformable image registration (DIR) for cervical cancer patients Purpose: Organ motion during radiation therapy in the pelvic region can potentially lead to uncertainties with the dose delivered to critical organs during fractionated treatment. The purpose of this study is to investigate, by means of using deformable image registration (DIR) and dose summation techniques, the differences between the planning dose and the delivered dose as calculated from the longitudinal cone-beam CT (CBCT) images for cervical cancer patients. Materials and Methods: Cervical cancer patients treated with external beam radiation therapy (EBRT) received a planning CT (pCT) and five CBCTs, once every five fractions of treatment. The “Merged CBCT” feature in MIM Maestro (MIM Software, Cleveland, OH) was performed between the pCT and each CBCT to generate an extended field-of-view (FOV) CBCT (mCBCT). A free-form multi-modality DIR was then performed between the pCT and the mCBCT to deform the pCT structures onto the mCBCT. DIR-generated bladder and rectum contours were further adjusted by a physician, and Dice Similarity Coefficients (DSC) were calculated between the two. After deformation, the investigated doses on the mCBCT were: 1) recalculated in Eclipse TPS (Varian Medical Systems, Palo Alto, CA) using original plan parameters (ecD), and 2) deformed from planning dose (pD) using the deformation matrix (mdD). Dose summation was performed to the first week’s mCBCT. Bladder D2cc, Dmax, Dmean, V45, and D50, rectum D2cc, Dmax, Dmean, and D50, and PTV45 D90 and D98 were compared between the three calculated doses. Dose distributions were compared in terms of dose volume histograms (DVHs) and gamma analysis. The Wilcoxon signed rank test was used to compare dosimetric metrics with statistical significance defined at p < 0.05. Results: For the ten patients analyzed, the average DSC were 0.72 ± 0.15 for bladder and 0.80 ± 0.11 for rectum. For most cases, only the superior and inferior slices were edited by physician. Regardless of the method of dose calculation (ecD or mdD), D2cc (bladder and rectum), and D90 and D98 (PTV45) were within 5% of pD for at least 9 out of 10 patients. For one patient each for bladder, rectum, and PTV45, the agreement was worse than 5%, with the largest difference of 15.3% for bladder D2cc in a patient with large bladder filling differences. For the Eclipse calculated dose on the merged CBCT (ecD) and t;he MIM deformed dose on merged CBCT (mdD), the bladder Dmax was within 5% for 8 out of 10 patients, and rectum Dmax was within 5% for 7 out of 10 patients. All 10 patients for ecD and mdD were >5% for bladder V45 due to the large variations in bladder volume throughout treatment. Statistically significant differences for bladder D2cc between the ecD and the mdD (p = 0.047). For bladder D50, significant differences between pD and ecD (p = 0.009) and ecD and mdD (p = 0.005). Statistically significant differences for rectum D2cc between the pD and ecD (p = 0.028) as well as ecD and mdD (p = 0.005). Statistically significant differences for D98 between the pD and ecD (p = 0.028) and pD and mdD (p = 0.007). The gamma analysis between the ecD and pD matched 90% of the voxels for 3 out of 10 patients and between the mdD and pD for 1 out of 10 patients. Conclusions: In this study, we evaluated cumulative doses based on weekly CBCTs using a commercially-available DIR software. Using DIR and the new Merged CBCT feature, we determined that reporting the initial planning dose would not introduce a more than 5% difference in 90% of cases studied. Our results indicate that the mdD produces similar dose values as the ecD for the OARs and PTV. The proposed workflow should be used on a case-by-case basis when the weekly CBCT shows marked difference in organs-at-risk from the planning CT.
Item Embargo Liver Vessel Segmentation and Accurate Landmark Pairs Detection for Quantitative Liver Deformable Image Registration Verification(2023) Zhang, ZhendongBackground: Target Registration Error (TRE) calculated based on selected anatomical landmarks is commonly known as the only trustable way to evaluate DIR accuracy. However, manual landmark pair selection is labor intensive and subjected to observer variability. Currently, no DIR benchmark datasets are available for Liver CTs. Manually selected landmark pairs have limited DIR evaluation power due to inadequate landmarks quantity (i.e., ~5 landmark pairs per dataset1 for liver CTs) and positional accuracy. For the purpose of liver DIR verification, there is a great need to establish a large quantity of landmark pairs with good positional accuracy.Purpose: An image processing procedure was developed in this study to automatically and precisely detect landmark pairs on corresponding vessel bifurcations between pairs of intra-patient CT images. With high positional accuracy, the generated landmark pairs can be used to evaluate deformable image registration (DIR) methods quantitatively for liver CTs. Methods: Landmark pairs were detected within the liver between pairs of contrast-enhanced CT scans for 32 patients. For each case, the liver vessel tree was automatically segmented in one image. Landmarks were automatically detected on vessel bifurcations. The corresponding landmarks in the second image were placed using a parametric DIR method (pTVreg). Manual validation was applied to reject outliers and adjust the landmarks’ positions to account for vessel segmentation uncertainty caused by the inconsistent image quality. Landmark pairs' positional accuracy of the procedure was evaluated using digital phantoms on target registration errors (TREs). Results: On average, ~71 landmark pairs per case were detected after manual outlier rejection. The proposed procedure increased the quantity of liver landmark pairs by ~10 times compared to the reported in the literature. A fully manual spot check showed that the reported procedure performed better than or as good as human at landmark pairs positional accuracy. Measured in the digital phantoms, the mean and standard deviation of TREs were 0.67 ± 0.48 mm with 99% of landmark pairs having TREs smaller than 2mm. Conclusion: A large number of liver landmark pairs with high positional accuracy were detected in contrasted enhanced CT image pairs using the reported method. The detected landmark pairs can be used for the quantitative evaluation of DIR methods.
Item Open Access Quantification of Lung Ventilation Using Voxel-based Delta Radiomics Extracted from Thoracic 4DCT(2020) Chen, XinruPurpose: Lung ventilation imaging offers guidance for functional avoidance during radiation therapy. Ventilation imaging modalities such as radioactive aerosols PET and SPECT, and hyperpolarized gas MRI, are not widely available at many institutions. In contrast, 4DCT images are part of standard treatment planning for lung malignancies and contain characteristics that reflect changes in the air content of the lungs due to ventilation. The purpose of this work was to develop a voxel-based delta radiomic feature extraction process using 4DCT images to quantify lung ventilation.
Materials and Methods: Twenty-five patients from the VAMPIRE dataset were used in this study with 4DCT/Galligas 4DPET images. For each patient, end-of-exhalation (EOE) and end-of-inhalation (EOI) phase CT images were both registered to the average phase CT using a contour-based deformable image registration algorithm. Next, 62 radiomic features were extracted spatially throughout the lungs using a sliding-window technique. The resulting tensor images were extracted to create 62 delta radiomic feature maps. Delta feature maps were compared with corresponding Galligas PET images by calculating Spearman correlation, mutual information. Delta feature distributions in clinical defect and non-defect lung regions were compared. The effect of sliding window kernel size was characterized to investigate its impact on correlation with Galligas PET.
Results: The best agreement between delta feature maps and Galligas PET images using a 5x5x5cm3 kernel was obtained by first-order energy, which demonstrates a mean Spearman correlation of r(s)=0.45±0.16. Other highly correlated filtered images were of features designed to capture high gray level intensities. Correlations with Galligas PET were found to increase and then saturate with increasing kernel size.
Conclusion: We have developed a promising method to quantify lung ventilation using voxel-based delta radiomics extracted from thoracic 4DCT. The results were comparable with a HU-based CT ventilation imaging (CTVI) method. Voxel-based radiomics is a potentially useful technique that can be used to generate synthetic ventilation images from standard-of-care image data.