Volume 35, Issue 6, June 2008
Index of content:
- Young Investigators Symposium: Auditorium C
- John R. Cameron: Young Investigators Symposium
SU‐HH‐AUD C‐01: Empirical Investigation of Flat‐Panel Imagers Incorporating High‐Efficiency, Segmented BGO and CsI:Tl Detectors for Radiotherapy Imaging35(2008); http://dx.doi.org/10.1118/1.2962308View Description Hide Description
Purpose: Megavoltage active matrix flat‐panel imagers (AMFPIs) are routinely used for portal imaging in external beam radiotherapy. However, conventional MV AMFPIs are inefficient, utilizing only ∼2% of the incident radiation, which leads to a detective quantum efficiency (DQE) of only ∼1%. Recent theoretical studies have shown that incorporation of thick, segmented scintillating detectors in MV AMFPIs can significantly increase DQE performance, leading to improved image quality at low dose and the possibility for dose‐efficient, cone‐beam computed tomography (CBCT).Method and Materials: Building upon the findings of our earlier studies, four prototype AMFPIs have recently been constructed, each consisting of a segmented BGO (11.3 mm thick) or CsI:Tl (11.4, 25.6 and 40.0 mm thick) detector coupled to an indirect detection flat‐panel array. Each detector consists of a matrix of 120 × 60 scintillator elements at 1.016 mm pitch, covered by a black or mirror top reflector. X‐ray sensitivity, MTF, NPS, DQE and phantom images were obtained for each prototype using a 6 MV photon beam at extremely low doses (e.g., 1 beam pulse, equivalent to 0.028 MU). Results: The BGO prototype shows better MTF (∼20% at the Nyquist frequency) than the CsI:Tl prototype at similar thickness. The measured DQE(0) at 1 beam pulse for the BGO and the three CsI:Tl prototypes were ∼19%, 12%, 19% and 23%, respectively. Moreover, the BGO prototype offers higher DQE values compared to the CsI:Tl prototypes over most of the spatial frequency range. Images of a contrast detail phantom using the BGO prototype at 2 beam pulses are comparable to that obtained from a conventional AMFPI with 18 times more dose.Conclusion: Prototype AMFPIs employing thick, segmented BGO and CsI:Tl detectors offer significant improvement in image quality at extremely low doses, leading to the possibility of soft tissue visualization using MV CBCT at clinically practical doses.
35(2008); http://dx.doi.org/10.1118/1.2962309View Description Hide Description
Purpose: The purpose of this study was to test the effectiveness of a new collimating material made of carbonfoam and low‐density polyethylene. The hypothesis is that the carbonfoam material will produce fewer neutrons and be more affordable to manufacture than conventional materials. The study includes comparison with spallations produced from a tungsten target and a combination between a new material and tungsten, as well as the effect of the products in Tissue Equivalent Plastic, and the energy deposition contribution to the total dose. The study also includes incorporating a new Multi‐leaf Collimator concept. Method and Materials The collimation material developed in this study is based on a space radiation shielding materials NASA project, and it is a low density Polyethylene that was impregnated into CarbonFoam (PELowFOAM). The material has a density of 1.071 g/cm3, a hydrogen atomic composition of 10.5% and a carbon atomic composition of 89.5%. Monte Carlo analyses for neutron and secondary protons production was performed (using HETC‐HEDS). Results: The number of neutrons produced with Tungsten is 65.78 % of the primary proton beam. PELowFoam and PELowFoam with tungsten produce almost 3.5 times fewer neutrons than tungsten alone. Secondary protons produced are not of concern. Almost 45.43 % of the primary beam deposited its energy as a scattering neutron into the Tissue equivalent Plastic. This number is alarming and needs to definitely be taken into consideration. Conclusion: A novel material was tested that is made of low density polyethylene impregnated in carbonfoam to be used for collimating proton beams instead of using high Z materials like tungsten. Out of the three materials, tungsten performed the worst producing 66% of the primary beam in the form of neutrons, while the PELowFoam produced about 16% neutrons, and the PELowFoam combined with tungsten produced about 19% neutrons.
SU‐HH‐AUD C‐03: Machine Learning Tools for Predicting Clinical Complications in a Multi‐Plan IMRT Framework35(2008); http://dx.doi.org/10.1118/1.2962310View Description Hide Description
Purpose: To apply machine learning algorithms to predict clinical outcomes (saliva flow rate, rectal bleeding, EUD, NTCP and PUC) for treatment plans taking into account dose and dose‐volume information. Method and Materials: One hundred twenty‐five (125) plans were generated for a representative head and neck case by varying the input dose‐volume constraints on the OARs (left parotid, right parotid and cord). The method that we used to predict continuous saliva flow rate from the treatment input constraint settings is a sequential minimal optimization algorithm for training a support vector regression model. Two hundred fifty‐six (256) plans were generated for the whole pelvis case by varying the input dose‐volume constraints on the rectum, bladder and bowel. We use the threshold of 66%/50Gy in conjunction with binary classification to predict rectal bleeding. The binary classification method employed is an optimized decision tree. We also calculated the EUD, NTCP, TCP and PUC for both 125 head and neck cases and 256 pelvic cases. Both Lyman‐Kutcher‐Burman (LKB) and Kallam and Agren‐Cronqvist's NTCP models were used. Both continuous and categorical values of these biological measures were predicted by modeling the planning surface using quadratic models and solving with linear programming with three different objective functions: minimizing maximum relative error, absolute error and sum of absolute errors. Results: We repeated the 10‐fold cross‐validation process 10 times and obtained a mean absolute error of 0.00078 with a 95% confidence interval [0.000769, 0.000792] for normalized saliva flow rate. We achieved an average correct prediction percentage of 97.35% and a 95% confidence interval for the correct prediction percentage as [96.89%, 97.81%]. Categorical correct prediction percentages for EUD, NTCP and PUC ranged from 75% to 96%. Conclusion: Our results indicate that machine learning techniques provide invaluable tools for the prediction of treatment complications from input constraints in a multi‐plan environment.
SU‐HH‐AUD C‐04: Improved Lesion Conspicuity in Dual‐Energy Imaging of the Chest: From NEQ to Observer Performance35(2008); http://dx.doi.org/10.1118/1.2962311View Description Hide Description
Purpose: To derive Fourier metrics of imaging performance (e.g., NEQ) in dual‐energy (DE) imaging of the chest that agree with human observer performance and to employ the resulting theoretical framework to system optimization with respect to lung nodule conspicuity. Method and Materials: The NEQ was computed using cascaded systems analysis extended to DE imaging and combined with a Fourier description of imaging task to yield an estimate of observer SNR (i.e., detectability index and AZ). Theoretical results were compared to human observer performance assessed in multiple alternative‐forced choice (MAFC) tests across a broad range of imaging conditions. The modeled observer SNR was used as an objective function for optimizing DE acquisition techniques and decomposition algorithms. A method for optimizing system performance for multiple imaging tasks was also investigated. Results:Theoretical calculations of the DE NEQ agreed well with measurements, and the task‐based detectability index was found to provide a strong predictor of human observer performance. Results identified [60/150] kVp as the optimal energy pair, with a weak dependence on high kVp. Optimization of the DE decomposition algorithms yielded significant improvements in lesion conspicuity — e.g., improving detection from barely visible (AZ <0.7) to highly conspicuous (AZ ∼1) at fixed dose to the patient. Optimal dose allocation (the fraction of total dose delivered in the low‐energy image) was found to range significantly — from 0.22 to 0.76 — depending on the choice of decomposition algorithm and imaging task. Conclusion: A theoretical model of DE imaging performance was derived and validated in comparison to human observers. The resulting framework provides a valuable guide to system optimization over a wide range of acquisition and decomposition conditions, yielding significant improvement in lung nodule conspicuity.
SU‐HH‐AUD C‐05: Low Dose Rate Prostate Brachytherapy: A Tomosynthesis‐Based Intra‐Operative Post‐Implant Dose Evaluation35(2008); http://dx.doi.org/10.1118/1.2962312View Description Hide Description
Purpose: To develop an intra‐operative dose assessment procedure that can be performed after an I‐125 prostate seed implantation, while the patient is still under anaesthesia. To accomplish this, we reconstruct the 3D position of each seed and co‐register it with the prostate contour. Method and materials: Our seed detection method involves a tomosynthesis‐based filtered reconstruction of the volume of interest. For 24 patients, the required cone‐beam images were obtained from 7 projections acquired over an angle of 60° with an isocentric imaging system adjacent to the treatment table. A graphical user interface (GUI) has been developed to allow visualization of the final seed positions and to interactively introduce corrections in the seeds positioning, if needed. The co‐registration between the tomosynthesis‐based seed positions and the TRUS‐based prostate contour is performed by applying the same rigid transformation as the one derived from the best match between the planned and the reconstructed seed positions. Doseanalysis is then performed based on the co‐registered images.Results: In a patient study with an average of 56 seeds per implant, the automatic tomosynthesis‐based reconstruction yields a detection rate of 96% of the seeds and less than 1 false‐positive seed per implant. The GUI allows the user to achieve a 100% detection rate in less than 5 minutes. The seed localization error obtained with a phantom study is (0.4±0.4) mm. This leads to small dosimetric relative errors on D90 and V100 of respectively 1.5% and 0.3%. Patient doseanalyses have shown a significant reduction in the dosimetric parameters between the planned and the post‐operative dosimetry. The relative difference between planned and intra‐operative D90 and V100 are respectively (11±8)% and (4±3)%. Conclusion: Our reconstruction method has the potential to provide accurate intra‐operative prostate dosimetry, all in less than 10 minutes extra time added to the whole implantation procedure.
35(2008); http://dx.doi.org/10.1118/1.2962313View Description Hide Description
Purpose: To integrate modern super‐resolution spectral quantification techniques with multi‐echo magnetic resonance imaging in order to provide multi‐parametric chemical shift imaging in real‐time. Method and Materials: A technique integrating the Steiglitz‐McBride algorithm and Cauchy's Calculus of Residues was developed to determine the proton resonance frequency(PRF), T2*, and T1‐wieghted amplitude for multiple species in a chemical shiftMR signal encoded using multiple gradient echoes. Monte Carlo simulation was implemented evaluating the performance of the algorithm under varying acquisition parameters and results compared to the Cramer‐Rao lower bound (CRLB). The PRF values in a mayonnaise‐lemon juice phantom were measured during heating demonstrating the ability of performing PRF thermometry in the presence of lipid. Phantom scans were performed comparing the T2* values from this technique and from spoiled gradient echo(SPGR) acquisitions. To demonstrate the use of this technique for temperature imaging in vivo, an image‐guided laser treatment in a canine brain was performed. Temperature maps were created at spatial resolutions of 1.5×1.5×4mm3 every five seconds in addition to T1‐weighted and T2* images providing multi‐parametric monitoring during treatment. Results: The technique achieves the CRLB demonstrating robust estimation of the PRF and T2* in the presence of noise. The uncertainty of the technique was theoretically and experimentally shown to be inversely proportional to the image SNR. The mayonnaise‐lemon juice phantom showed a PRF shift with temperature consistent to published literature. T2* values provided by the technique did not statistically differ from SPGR‐based T2* mapping. Monitoring of the canine brain treatment was successful providing highly sensitive temperature measurements. T2*‐corrected T1‐W images detected a treatment‐induced hemorrhage demonstrating the importance of multi‐parametric monitoring. Conclusion: A technique is presented that provides chemical shift imaging at higher spatiotemporal resolutions than what was previously available. The ability of this technique to provide multiple parameters demonstrates great promise for image‐guided thermal therapies.
SU‐HH‐AUD C‐07: Classification of Breast Carcinoma Subtypes Using Computer‐Extracted Morphological and Kinetic Features in DCE‐MRI35(2008); http://dx.doi.org/10.1118/1.2962314View Description Hide Description
Purpose: To assess the performance of computer‐extracted morphological and kinetic features in differentiating different types of breast lesions, specifically ductal carcinoma in situ (DCIS) and invasive ductal carcinoma (IDC), in DCE‐MRI. Method and Materials: Breast MR images were obtained with a T1‐weighted SPGR sequence using Gd‐DTPA on a 1.5T GEMRI scanner. Each case has one precontrast and five postcontrast series at intervals of 68 seconds, and each series contains 60 coronal slices. The database contains 83 benign breast lesions and 98 malignant breast lesions including 24 DCIS, 32 IDC, 18 combined DCIS/IDC, 10 invasive lobular carcinoma (ILC) and 14 non‐categorizable lesions. All lesions were verified pathologically. Only pure DCIS and IDC lesions were used. Each lesion was segmented and its characteristic kinetic curve was extracted using the fuzzy c‐means method. Twenty‐nine features including textural, morphological, kinetic, and variance kinetic features were extracted and stepwise linear discriminant analysis using a Wilks lambda cost function was used for feature selection. The selected features were merged using round‐robin Bayesianneural network and the classification performance was evaluated using receiver‐operating characteristics (ROC) analysis. Results: Distinguishing DCIS lesions from IDC lesions using selected features gave an AUC value of 0.95. AUC values of 0.85 and 0.87 were obtained in differentiating between DCIS and benign lesions, and between IDC and benign lesions with merged features, respectively. Conclusion: Computer‐aided diagnosis for breast MRI can be extended from malignant vs. benign classification to distinguishing different subcategories of breast carcinomas, in particular DCIS and IDC.
SU‐HH‐AUD C‐08: Spatial Weighted Mutual Information for Image Registration in Image Guided Radiation Therapy35(2008); http://dx.doi.org/10.1118/1.2962315View Description Hide Description
Purpose: To develop a robust registration algorithm for medical images, emphasizing the registration of plan images with cone beam computed tomography(CBCT)images for IGRT. Our Spatial Weighted Mutual Information (SWMI) technique assigns greater importance to user‐selected volume, allowing medically ‘important’ areas to dominate the registration process. Method and Materials: Mutual Information (MI) is the most popular measure of image registration due to its robustness and multi‐modal capability. However, MI does have difficulty where organ deformation is present. We reformulated the MI algorithm by incorporating an adaptable weight function to user selected spatial locations. Since MI is defined in probabilistic space, we proposed a spatial‐weighted joint probability, and composed a Spatial Weighted Mutual Information measure. Our image registrationsoftware and Graphical User Interface (GUI) was programmed in C++ to import DICOM images and DICOM‐RT information, and to perform image registration. For this study, we used Gaussian‐shaped spatial weights applied to a user‐defined volume. Our software allows a user to adjust the Gaussian parameters via the GUI. Convergence and robustness of our registration method was first tested with a head‐and‐neck plan and seven CBCTimage sets. Then, a prostate plan with nine CBCTimage sets was analyzed. Speed of convergence was tested by arbitrarily miss‐aligning two image sets ±15mm over 41 trials. We also applied our algorithm to fuse CT and MRI image sets. Results and Discussion:Image registration using our measure converges 10% faster than using Mutual Information. Our study showed image registration using a uniform weight over an entire volume lead to compromised target coverage. SWMI showed better alignment near target areas and neighboring critical organs even with organ deformation. Our method worked well in fusing MRI to CTimages as well.
(This work is partly supported by Susan G. Komen Breast Foundation Grant: BTCR126506).
SU‐HH‐AUD C‐09: Bioanatomic MR Imaging for Characterization of Brain Tumor and Radiation Response in the Rat Brain35(2008); http://dx.doi.org/10.1118/1.2962316View Description Hide Description
Purpose: To examine the efficacy of bioanatomic magnetic resonance imaging(MRI) in braintumor characterization, and the implications for radiation therapy (RT) treatment planning and assessment of treatment response. Method and Materials:Braintumor models were created by implanting human high‐grade glioma cells into immunodeficient rats. Once the tumors reached a specified target size, animals underwent Gamma Knife™ Stereotactic Radiosurgery (GK SRS, 15–18 Gy to the 50 % isodose line). During tumor growth and after irradiation, animals underwent serial MR evaluation (dynamic susceptibility contrast, DSC‐MRI) to quantify changes in cerebral blood volume (CBV) throughout the brain regions containing the tumor. Upon completion of the experiment, CBV data were compared with conventional MRimaging data and histology to evaluate the usefulness of the additional biological information with respect to target volume delineation and response to irradiation. Results: A DSC‐MRI erfusion protocol was successfully developed to accurately quantify blood volume changes within the brain. During tumor progression, active areas of tumor growth showed increased CBV, while areas of developing necrosis demonstrated decreased CBV. Large intra‐tumor CBV heterogeneities were visible in the late stages of unirradiated tumors and within a week of those irradiated with GK SRS. In both cases CBV data provided biological information regarding tumor behavior, which was not elucidated by the corresponding anatomical images.Conclusion: This work illustrates the potential of bioanatomic MRI for improving braintumorradiation therapy. Most current high‐grade glioma RT techniques focus solely on anatomical images, which often neglect biological information when determining target volumes. By incorporating DSC‐MRI into the therapy planning and response assessment procedures, changes in CBV could be used to better understand tumor behavior. This could eventually lead to a more accurate way to determine conformal target volumes, and a more sensitive method for monitoring the therapeutic response.
SU‐HH‐AUD C‐10: The Imaging and Dosimetric Capabilities of a Novel CT/MR‐Suitable, Anatomically Adaptive, Shielded HDR/PDR Intracavitary Brachytherapy Applicator for the Treatment of Cervical Cancer35(2008); http://dx.doi.org/10.1118/1.2962317View Description Hide Description
Purpose: To design and investigate the imaging and dosimetric capabilities of a novel, CT/MR‐suitable, anatomically adaptive, shielded cervical HDR/PDR brachytherapy applicator. Method and Materials: An applicator was constructed featuring an inter‐colpostat shield that can translate/rotate about the colpostat's long‐axis. Artifact‐free CTimaging was achieved using a “step‐and‐shoot” technique; pausing the scanner midway through the scan and moving the shield out of the beam's path. Artifact‐free MRIimaging was achieved by utilizing MRI‐compatible ovoid components and pulse‐sequences that minimize susceptibility artifacts. The applicator's imaging capabilities were demonstrated acquiring images using phantoms that positioned the novel and Fletcher‐Williamson ICBT applicators in clinically‐applicable geometries for both modalities. Artifacts were qualitatively compared. To evaluate any dosimetric advantages, Monte‐Carlo models of the novel and FW applicators were first validated. Anatomies of patients that have undergone ICBT for cervical disease were modeled using Monte‐Carlo and spatially registered with models of both applicators using SolidWorks; a CAD software suite. Equivalent, clinically‐applicable loadings were simulated for both applicators using Monte‐Carlo techniques. The novel applicator shield's rotation and translation was adjusted for each dwell position in order to minimize dose to the rectum and superimposed for comparison to equivalent FW treatments. Rectal dose (rate and absolute) volume histograms were determined for both applicators and compared. Results: Using a “step‐and‐shoot” CT scanning method and MR compliant materials and optimal pulse‐sequences, images of the novel applicator were artifact‐free using both modalities. Additionally, for the patient case presented, there is a 26% and 13% reduction of d90 and d50, respectively, and a 13% reduction in overall absolute dose to the rectum when compared to equivalent Fletcher‐Williamson ICBT treatments. Conclusion: A novel ICBT applicator can be imaged using CT/MR without artifact and reduce dose to the rectum compared to the current state‐of‐the‐art, FW ICBT applicator. Conflict‐of‐Interest: Work partially supported by Nucletron Corp.