J Xia

High performance computing for deformable image registration: towards a new paradigm in adaptive radiotherapy.

The advent of readily available temporal imaging or time series volumetric (4D) imaging has become an indispensable component of treatment planning and adaptive radiotherapy (ART) at many radiotherapy centers. Deformable image registration (DIR) is also used in other areas of medical imaging, including motion corrected image reconstruction. Due to long computation time, clinical applications of DIR in radiation therapy and elsewhere have been limited and consequently relegated to offline analysis. With the recent advances in hardware and software, graphics processing unit (GPU) based computing is an emerging technology for general purpose computation, including DIR, and is suitable for highly parallelized computing. However, traditional general purpose computation on the GPU is limited because the constraints of the available programming platforms. As well, compared to CPU programming, the GPU currently has reduced dedicated processor memory, which can limit the useful working data set for parallelized processing. We present an implementation of the demons algorithm using the NVIDIA 8800 GTX GPU and the new CUDA programming language. The GPU performance will be compared with single threading and multithreading CPU implementations on an Intel dual core 2.4GHz CPU using the C programming language. CUDA provides a C-like language programming interface, and allows for direct access to the highly parallel compute units in the GPU. Comparisons for volumetric clinical lung images acquired using 4DCT were carried out. Computation time for 100 iterations in the range of 1.8-13.5s was observed for the GPU with image size ranging from 2.0×106to14.2×106pixels. The GPU registration was 55-61 times faster than the CPU for the single threading implementation, and 34-39 times faster for the multithreading implementation. For CPU based computing, the computational time generally has a linear dependence on image size for medical imaging data. Computational efficiency is characterized in terms of time per megapixels per iteration (TPMI) with units of seconds per megapixels per iteration (or spmi). For the demons algorithm, our CPU implementation yielded largely invariant values of TPMI. The mean TPMIs were 0.527spmi and 0.335spmi for the single threading and multithreading cases, respectively, with <2% variation over the considered image data range. For GPU computing, we achieved TPMI=0.00916 spmi with 3.7% variation, indicating optimized memory handling under CUDA. The paradigm of GPU based real-time DIR opens up a host of clinical applications for medical imaging.

Fast Deformable Registration on the GPU: A CUDA Implementation of Demons

In the medical imaging field, we need fast deformable registration methods especially in intra-operative settings characterized by their time-critical applications. Image registration studies which are based on graphics processing units (GPUs) provide fast implementations. However, only a small number of these GPU-based studies concentrate on deformable registration. We implemented Demons, a widely used deformable image registration algorithm, on NVIDIApsilas Quadro FX 5600 GPU with the compute unified device architecture (CUDA) programming environment. Using our code, we registered 3D CT lung images of patients. Our results show that we achieved the fastest runtime among the available GPU-based Demons implementations. Additionally, regardless of the given dataset size, we provided a factor of 55 speedup over an optimized CPU-based implementation. Hence, this study addresses the need for on-line deformable registration methods in intra-operative settings by providing the fastest and most scalable Demons implementation available to date. In addition, it provides an implementation of a deformable registration algorithm on a GPU, an understudied type of registration in the general-purpose computation on graphics processors (GPGPU) community.

A real‐time respiratory motion monitoring system using KINECT: Proof of concept

PURPOSE The purpose of this study is to investigate the feasibility of a low-cost respiratory motion monitoring system based on the Microsoft KINECT sensor. METHODS The authors increased KINECTs inherent depth resolution from 1 cm to 1 mm via a motion magnification system. Using the KINECT software development kit, the authors programmed the KINECT to capture depth images and determine the average depth over a thoracic region of interest, viewed almost parallel to the subjects surface. KINECT respiratory traces (average depth vs time at a rate of 30 Hz) were acquired from four volunteers and compared with those simultaneously acquired using a commercially available strain gauge respiratory gating system. RESULTS The correlation coefficient (CC) between KINECT and strain gauge traces varied from 0.958 to 0.978, with a mean CC of 0.969. This strong correlation was also demonstrated by the joint probability distribution and visual inspection. CONCLUSIONS It is feasible to use the KINECT for respiratory motion tracking. Traces are similar to those of a clinically used strain gauge system. The KINECT-based system provides a new and economical way to monitor respiratory motion.

High resolution (3 Tesla) MRI-guided conformal brachytherapy for cervical cancer: consequences of different high-risk CTV sizes

Purpose To evaluate conventional brachytherapy (BT) plans using dose-volume parameters and high resolution (3 Tesla) MRI datasets, and to quantify dosimetric benefits and limitations when MRI-guided, conformal BT (MRIG-CBT) plans are generated. Material and methods Fifty-five clinical high-dose-rate BT plans from 14 cervical cancer patients were retrospectively studied. All conventional plans were created using MRI with titanium tandem-and-ovoid applicator (T&O) for delivery. For each conventional plan, a MRIG-CBT plan was retrospectively generated using hybrid inverse optimization. Three categories of high risk (HR)-CTV were considered based on volume: non-bulky (< 20 cc), low-bulky (> 20 cc and < 40 cc) and bulky (≥ 40 cc). Dose-volume metrics of D90 of HR-CTV and D2cc and D0.1cc of rectum, bladder, and sigmoid colon were analyzed. Results Tumor coverage (HR-CTV D90) of the conventional plans was considerably affected by the HR-CTV size. Sixteen percent of the plans covered HR-CTV D90 with the prescription dose within 5%. At least one OAR had D2cc values over the GEC-ESTRO recommended limits in 52.7% of the conventional plans. MRIG-CBT plans showed improved target coverage for HR-CTV D90 of 98 and 97% of the prescribed dose for non-bulky and low-bulky tumors, respectively. No MRIG-CBT plans surpassed the D2cc limits of any OAR. Only small improvements (D90 of 80%) were found for large targets (> 40 cc) when using T&O applicator approach. Conclusions MRIG-CBT plans displayed considerable improvement for tumor coverage and OAR sparing over conventional treatment. When the HR-CTV volume exceeded 40 cc, its improvements were diminished when using a conventional intracavitary applicator.

Impact of temporal probability in 4D dose calculation for lung tumors.

The purpose of this study was to evaluate the dosimetric uncertainty in 4D dose calculation using three temporal probability distributions: uniform distribution, sinusoidal distribution, and patient‐specific distribution derived from the patient respiratory trace. Temporal probability, defined as the fraction of time a patient spends in each respiratory amplitude, was evaluated in nine lung cancer patients. Four‐dimensional computed tomography (4D CT), along with deformable image registration, was used to compute 4D dose incorporating the patients respiratory motion. First, the dose of each of 10 phase CTs was computed using the same planning parameters as those used in 3D treatment planning based on the breath‐hold CT. Next, deformable image registration was used to deform the dose of each phase CT to the breath‐hold CT using the deformation map between the phase CT and the breath‐hold CT. Finally, the 4D dose was computed by summing the deformed phase doses using their corresponding temporal probabilities. In this study, 4D dose calculated from the patient‐specific temporal probability distribution was used as the ground truth. The dosimetric evaluation matrix included: 1) 3D gamma analysis, 2) mean tumor dose (MTD), 3) mean lung dose (MLD), and 4) lung V20. For seven out of nine patients, both uniform and sinusoidal temporal probability dose distributions were found to have an average gamma passing rate >95% for both the lung and PTV regions. Compared with 4D dose calculated using the patient respiratory trace, doses using uniform and sinusoidal distribution showed a percentage difference on average of −0.1%±0.6% and −0.2%±0.4% in MTD, −0.2%±1.9% and −0.2%±1.3% in MLD, 0.09%±2.8% and −0.07%±1.8% in lung V20, −0.1%±2.0% and 0.08%±1.34% in lung V10, 0.47%±1.8% and 0.19%±1.3% in lung V5, respectively. We concluded that four‐dimensional dose computed using either a uniform or sinusoidal temporal probability distribution can approximate four‐dimensional dose computed using the patient‐specific respiratory trace. PACS number: 87.55.D‐The purpose of this study was to evaluate the dosimetric uncertainty in 4D dose calculation using three temporal probability distributions: uniform distribution, sinusoidal distribution, and patient-specific distribution derived from the patient respiratory trace. Temporal probability, defined as the fraction of time a patient spends in each respiratory amplitude, was evaluated in nine lung cancer patients. Four-dimensional computed tomography (4D CT), along with deformable image registration, was used to compute 4D dose incorporating the patients respiratory motion. First, the dose of each of 10 phase CTs was computed using the same planning parameters as those used in 3D treatment planning based on the breath-hold CT. Next, deformable image registration was used to deform the dose of each phase CT to the breath-hold CT using the deformation map between the phase CT and the breath-hold CT. Finally, the 4D dose was computed by summing the deformed phase doses using their corresponding temporal probabilities. In this study, 4D dose calculated from the patient-specific temporal probability distribution was used as the ground truth. The dosimetric evaluation matrix included: 1) 3D gamma analysis, 2) mean tumor dose (MTD), 3) mean lung dose (MLD), and 4) lung V20. For seven out of nine patients, both uniform and sinusoidal temporal probability dose distributions were found to have an average gamma passing rate >95% for both the lung and PTV regions. Compared with 4D dose calculated using the patient respiratory trace, doses using uniform and sinusoidal distribution showed a percentage difference on average of -0.1%±0.6% and -0.2%±0.4% in MTD, -0.2%±1.9% and -0.2%±1.3% in MLD, 0.09%±2.8% and -0.07%±1.8% in lung V20, -0.1%±2.0% and 0.08%±1.34% in lung V10, 0.47%±1.8% and 0.19%±1.3% in lung V5, respectively. We concluded that four-dimensional dose computed using either a uniform or sinusoidal temporal probability distribution can approximate four-dimensional dose computed using the patient-specific respiratory trace. PACS number: 87.55.D.

SU‐E‐J‐260: Dose Recomputation Versus Dose Deformation for Stereotactic Body Radiation Therapy in Lung Tumors: A Dosimetric Study

PURPOSE To evaluate the dosimetric accuracy between recomputed dose and deformed dose for stereotactic body radiation therapy in lung tumors. METHODS Two non-small-cell lung cancer patients were analyzed in this study, both of whom underwent 4D-CT and breath-hold CT imaging. Treatment planning was performed using the breath-hold CT images for the dose calculation and the 4D-CT images for determining internal target volumes. 4D-CT images were reconstructed with ten breathing amplitude for each patient. Maximum tumor motion was 13 mm for patient 1, and 7 mm for patient 2. The delivered dose was calculated using the 4D-CT images and using the same planning parameters as for the breath-hold CT. The deformed dose was computed by deforming the planning dose using the deformable image registration between each binned CT and the breath-hold CT. RESULTS For patient 1, the difference between recomputed dose and deformed mean lung dose (MLD) ranged from 11.3%(0.5 Gy) to 1.1%(0.06 Gy), mean tumor dose (MTD) ranged from 0.4%(0.19 Gy) to -1.3%(-0.6 Gy), lung V20 ranged from +0.74% to -0.33%. The differences in all three dosimetric criteria remain relatively invariant to target motion. For patient 2, V20 ranged from +0.42% to -2.41%, MLD ranged from -0.2%(-0.05 Gy) to -10.4%(-2.12 Gy), and MTD ranged from -0.5%(-0.31 Gy) to -5.3%(-3.24 Gy). The difference between recomputed dose and deformed dose shows strong correlation with tumor motion in all three dosimetric measurements. CONCLUSION The correlation between dosimetric criteria and tumor motion is patient-specific, depending on the tumor locations, motion pattern, and deformable image registration accuracy. Deformed dose can be a good approximation for recalculated dose when tumor motion is small. This research is supported by Siemens Medical Solutions USA, Inc and Iowa Center for Research By Undergraduates.

Development of a radiobiological evaluation tool to assess the expected clinical impacts of contouring accuracy between manual and semi-automated segmentation algorithms

RADEval is a tool developed to assess the expected clinical impact of contouring accuracy when comparing manual contouring and semi-automated segmentation. The RADEval tool, designed to process large scale datasets, imported a total of 2,760 segmentation datasets, along with a Simultaneous Truth and Performance Level Estimation (STAPLE) to act as ground truth tumor segmentations. Virtual dose-maps were created within RADEval and two different tumor control probability (TCP) values using a Logistic and a Poisson TCP models were calculated in RADEval using each STAPLE and each dose-map. RADEval also virtually generated a ring of normal tissue. To evaluate clinical impact, two different uncomplicated TCP (UTCP) values were calculated in RADEval by using two TCP-NTCP correlation parameters (δ = 0 and 1). NTCP values showed that semi-automatic segmentation resulted in lower NTCP with an average 1.5 – 1.6 % regardless of STAPLE design. This was true even though each normal tissue was created from each STAPLE (p < 0.00001). TCP and UTCP presented no statistically significant differences (p ≥ 0.1884). The intra-operator standard deviations (SDs) for TCP, NTCP and UTCP were significantly lower for the semi-automatic segmentation method regardless of STAPLE design (p < 0.0331). Both intra-and inter-operator SDs of TCP, NTCP and UTCP were significantly lower for semi-automatic segmentation for the STAPLE 1 design (p <0.0331). RADEval was able to efficiently process 4,920 datasets of two STAPLE designs and successfully assess the expected clinical impact of contouring accuracy.

WE-DE-BRA-10: Development of a Novel Scanning Beam Low-Energy Intraoperative Radiation Therapy (SBIORT) System for Pancreatic Cancer

PURPOSE Developing a compact collimator system and validating a 3D surface imaging module for a scanning beam low-energy x-ray radiation therapy (SBIORT) system that enables delivery of non-uniform radiation dose to targets with irregular shapes intraoperatively. METHODS SBIORT consists of a low energy x-ray source, a custom compact collimator module, a robotic arm, and a 3D surface imaging module. The 3D surface imaging system (structure sensor) is utilized for treatment planning and motion monitoring of the surgical cavity. SBIORT can deliver non-uniform dose distributions by dynamically moving the x-ray source assembly along optimal paths with various collimator apertures. The compact collimator utilizes a dynamic shutter mechanism to form a variable square aperture. The accuracy and reproducibility of the collimator were evaluated using a high accuracy encoder and a high resolution camera platform. The dosimetrical characteristics of the collimator prototype were evaluated using EBT3 films with a Pantak Therapax unit. The accuracy and clinical feasibility of the 3D imaging system were evaluated using a phantom and a cadaver cavity. RESULTS The SBIORT collimator has a compact size: 66 mm diameter and 10 mm thickness with the maximum aperture of 20 mm. The mechanical experiment indicated the average accuracy of leaf position was 0.08 mm with a reproducibility of 0.25 mm at 95% confidence level. The dosimetry study indicated the collimator had a penumbra of 0.35 mm with a leaf transmission of 0.5%. 3D surface scans can be acquired in 5 seconds. The average difference between the acquired 3D surface and the ground truth is 1 mm with a standard deviation of 0.6 mm. CONCLUSION This work demonstrates the feasibility of the compact collimator and 3D scanning system for the SBIORT. SBIORT is a way of delivering IORT with a compact system that requires minimum shielding of the procedure room. This research is supported by the University of Iowa Internal Funding Initiatives.

Clinical validation of a real-time applicator position monitoring system for gynecologic intracavitary brachytherapy

Purpose: To validate the clinical feasibility and efficacy of a real-time applicator position monitoring system (RAPS) through a phantom study and a prospective clinical trial. Methods and materials: The RAPS measures the brachytherapy applicator displacement in real-time by computing the relative displacement between two infrared reflective targets, one attached to the applicator and the other to the patients skin. A phantom study was performed to compare RAPS measurements with the ground truth. Six cervical cancer patients were enrolled in the clinical trial using MRI-based high-dose-rate brachytherapy with a Tandem-and-Ovoids applicator. The results from the RAPS are compared with the clinical method. Results: In the phantom study, an average difference between RAPS measurements and known displacements was 0.02 ± 0.01 mm in the superior-inferior direction, 0.02 ± 0.02 mm in the lateral direction, and 0.11 ± 0.06 mm in the anterior-posterior direction. In the clinical trial, the absolute difference in applicator displacement between the RAPS and the clinical method was 1.46 ± 1.13 mm. In all patient cases, a maximum applicator displacement of 6.66 mm (2.0 ± 1.5 mm) was observed using the RAPS. Conclusions: This work demonstrates the clinical efficacy of RAPS to measure applicator displacement.

SU‐E‐J‐200: A Dosimetric Analysis of 3D Versus 4D Image‐Based Dose Calculation for Stereotactic Body Radiation Therapy in Lung Tumors

PURPOSE To evaluate the dosimetric difference between 3D and 4Dweighted dose calculation using patient specific respiratory trace and deformable image registration for stereotactic body radiation therapy in lung tumors. METHODS Two dose calculation techniques, 3D and 4D-weighed dose calculation, were used for dosimetric comparison for 9 lung cancer patients. The magnitude of the tumor motion varied from 3 mm to 23 mm. Breath-hold exhale CT was used for 3D dose calculation with ITV generated from the motion observed from 4D-CT. For 4D-weighted calculation, dose of each binned CT image from the ten breathing amplitudes was first recomputed using the same planning parameters as those used in the 3D calculation. The dose distribution of each binned CT was mapped to the breath-hold CT using deformable image registration. The 4D-weighted dose was computed by summing the deformed doses with the temporal probabilities calculated from their corresponding respiratory traces. Dosimetric evaluation criteria includes lung V20, mean lung dose, and mean tumor dose. RESULTS Comparing with 3D calculation, lung V20, mean lung dose, and mean tumor dose using 4D-weighted dose calculation were changed by -0.67% ± 2.13%, -4.11% ± 6.94% (-0.36 Gy ± 0.87 Gy), -1.16% ± 1.36%(-0.73 Gy ± 0.85 Gy) accordingly. CONCLUSION This work demonstrates that conventional 3D dose calculation method may overestimate the lung V20, MLD, and MTD. The absolute difference between 3D and 4D-weighted dose calculation in lung tumor may not be clinically significant. This research is supported by Siemens Medical Solutions USA, Inc and Iowa Center for Research By Undergraduates.