Browsing by Department "Medical Physics DKU"
Results Per Page
Sort Options
Item Open Access A novel mono-energy proton arc therapy with patient specific range shifter for fast treatment delivery(2024) Zhou, YuyinAbstractIntroduction:This study evaluates a new proton therapy filter designed to eliminate the need for energy adjustments. Utilizing the machine's maximum energy, the filter ensures sufficient tumor coverage through the Bragg-peak, potentially improving treatment efficiency by shortening delivery time. Methods:Implemented on the matRad platform, each plan utilized a single arc composed of 72 beams, each spaced 5 degrees apart. Open-access datasets, including TG-119 C-shape, a prostate case, and a liver case, were employed. The prescribed doses for these cases were 50Gy in 25 fractions, 68Gy in 34 fractions, and 45Gy in 25 fractions, respectively. Simplifying from multiple energy layers to a single energy layer for each beam can reduce treatment delivery time. Maintaining spot coverage with a single energy layer for each beam is a critical optimization aspect. The spot coverage, P(i,j), is optimized to maximize spot coverage and the optimization is called mono energy optimization. However, considering spot coverage alone is insufficient; the energy level must also be considered. Higher energy levels indicate a thinner range shifter, which reduces scatter and attenuation caused by range shifters. The new optimization process, called higher mono energy optimization, gave priority to deeper layers and larger spot sizes, using a function that normalizes input energy and combines it with alpha and beta coefficients to optimize the energy function E(i,j) and spot coverage P(i,j). The optimal energy layers were selected, and the initial beam energy was set at 236MeV. All beamlet was adjusted to specific energy levels with a custom-designed PMMA filter based on stopping power, facilitating a smooth transition to the desired energy levels. The effectiveness of this approach was evaluated by comparing dose metrics with those from the Intensity Modulated Proton Therapy (IMPT) method using two or three beams. Results: PTV coverages were relatively close between the IMPT and range filter plans. Organs at Risk (OAR) experienced a dose increase due to enhanced scattering. Simulated treatment delivery times for the three tested range filter plans demonstrated the efficiency, with prostate at 360s, liver at 340s, and TG119 at 390s. Conclusions: Mono-energy with range filters proton therapy is a feasible approach for expediting treatment delivery without compromising the quality of the treatment plan.
Item Open Access A Radiomics Machine Learning Model for Post-Radiotherapy Overall Survival Prediction of Non-Small Cell Lung Cancer (NSCLC)(2023) Zhang, RihuiPurpose: To predict post-radiotherapy overall survival group of NSCLC patients based on clinical information and radiomics analysis of simulation CT. Materials/Methods: A total of 258 non-adenocarcinoma patients who received radical radiotherapy or chemo-radiation were studied: 45/50/163 patients were identified as short(0-6mos)/mid(6-12mos)/long(12+mos) survival groups, respectively. For each patient, we first extracted 76 radiomics features within the gross tumor volume(GTV) identified in the simulation CT; these features were combined with patient clinical information (age, overall stage, and GTV volume) as a patient-specific feature vector, which was utilized by a 2-step machine learning model for survival group prediction. This model first identifies patients with long survival prediction via a supervised binary classifier; for those with otherwise prediction, a 2nd classifier further generates short/mid survival prediction. Two machine learning classifiers, explainable boosting machine(EBM) and balanced random forest(BRF), were interrogated as a comparison study. During the model training, all patients were divided into training/test sets by an 8:2 ratio, and 100-fold random sampling were applied to the training set with a 7:1 validation ratio. Model performances were evaluated by the sensitivity, accuracy, and ROC results. Results: The model with EBM demonstrated an overall ROC AUC (0.58±0.04) with limited sensitivities in short (0.02±0.04) and mid group (0.11±0.08) predictions due to imbalanced data sample distribution. In contrast, the model with BRF improved short/mid group sensitivities to 0.32±0.11/0.29±0.16, respectively, but the improvement of ROC AUC (0.60±0.04) is limited. Nevertheless, both EBM (0.46±0.04) and BRF (0.57±0.04) approaches achieved limited overall accuracy; a noticeable overlap was found in their feature lists with top 10 feature weight rankings. Conclusion: The proposed two-step machine learning model with BRF classifier possesses a better performance than the one with EBM classifier in the post-radiotherapy survival group prediction of NSCLC. Future works, preferably in the joint use of deep learning, are in demand to further improve the prediction results.
Item Open Access A Radiomics-Incorporated Deep Ensemble Learning Model for Multi-Parametric MRI-based Glioma Segmentation(2023) YANG, CHENAbstractPurpose: To develop a deep ensemble learning model with a radiomics spatial encoding execution for improved glioma segmentation accuracy using multi-parametric MRI (mp-MRI). Materials/Methods: This radiomics-incorporated deep ensemble learning model was developed using 369 glioma patients with a 4-modality mp-MRI protocol: T1, contrast-enhanced T1 (T1-Ce), T2, and FLAIR. In each modality volume, a 3D sliding kernel was implemented across the brain to capture image heterogeneity: fifty-six radiomic features were extracted within the kernel, resulting in a 4th order tensor. Each radiomic feature can then be encoded as a 3D image volume, namely a radiomic feature map (RFM). For each patient, all RFMs extracted from all 4 modalities were processed by the Principal Component Analysis (PCA) for dimension reduction, and the first 4 principal components (PCs) were selected. Next, four deep neural networks following the U-net’s architecture were trained for the segmenting of a region-of-interest (ROI): each network utilizes the mp-MRI and 1 of the 4 PCs as a 5-channel input for 2D execution. Last, the 4 softmax probability results given by the U-net ensemble were superimposed and binarized by Otsu’s method as the segmentation result. Three deep ensemble models were trained to segment enhancing tumor (ET), tumor core (TC), and whole tumor (WT), respectively. Segmentation results given by the proposed ensemble were compared to the mp-MRI-only U-net results. Results: All 3 radiomics-incorporated deep learning ensemble models were successfully implemented: Compared to mp-MRI-only U-net results, the dice coefficients of ET (0.777→0.817), TC (0.742→0.757), and WT (0.823→0.854) demonstrated improvements. Accuracy, sensitivity, and specificity results demonstrated the same patterns. Conclusion: The adopted radiomics spatial encoding execution enriches the image heterogeneity information that leads to the successful demonstration of the proposed neural network ensemble design, which offers a new tool for mp-MRI-based medical image segmentation.
Item Open Access Application of TG-218 to SRS and SBRT Pre-Treatment Patient Specific QA(2020) Xia, YuqingAbstract
Purpose: Updated recommendations for pre-treatment QA of patient-specific intensity modulated radiation therapy (IMRT) and Volumetric modulated arc therapy (VMAT) quality assurance (QA) were recently published by the AAPM task group TG-218. While the traditionally most common QA analysis is to use a Gamma index with dose & spatial analysis criteria of 3% & 3mm, respectively, TG-218 recommends a tighter spatial tolerance of 2mm for standard IMRT QA, and that even tighter tolerances should be considered for stereotactic radiosurgery (SRS) and stereotactic body radiotherapy (SBRT). Our purpose is to report our experience with applying the TG-218 recommendations to a large clinical SRS and SBRT program. In addition, a new SRS technique was recently developed at Duke, called Conformal Arc Informed Volumetric Modulated Arc Therapy (CAVMAT), which is designed to be less sensitive to configuration and delivery errors. We measured the agreement of CAVMAT for pre-treatment QA and compared it to the current standard (VMAT) to evaluate whether CAVMAT is more robust to delivery errors than VMAT.
Methods: We re-analyzed the pre-treatment QA with respect to the TG-218 recommendations. For Portal Dosimetry (Varian Medical Systems, Palo Alto, CA), this included IMRT brain (n=25) and SBRT / hypofractionated image guided radiotherapy (HIGRT) cases that utilize flattened photon beams (n=18). For Delta4 (ScandiDos, Madison, WI) this included single target SRS (n=24), multiple target SRS (n=25), SBRT cases using VMAT (n=74), and SBRT cases using IMRT with FFF photons (n=23). For ArcCHECK (Sun Nuclear, Melbourne, FL)), we take 25 single target VMAT SRS cases and 25 multiple target VMAT SRS cases. For SRS MapCHECK(Sun Nuclear, Melboume, FL), we analyze 10 multiple target VMAT SRS cases with 16 targets. A Gamma analysis was performed with 6 spatial/dose criteria combinations: 3%/3mm, 3%/2mm, 3%/1mm, 2%/1mm, 4%/1mm, 5%/1mm. We then calculated the TG-218 action limit and tolerance limit per plan type and compared to the “universal” TG-218 action limit of 90% having a Gamma <1.
To compare CAVMAT and VMAT, log file analysis and pre-treatment QA was performed for 10 patients with 20 plans (10 VMAT, 10 CAVMAT) with 46 targets in total. 10 VMAT plans were re-planned using CAVMAT, and the dosimetric effect due to treatment delivery errors was quantified for V6Gy, V12Gy, and V16Gy of healthy brain along with the maximum, average and minimum doses of each target. Gamma analysis of VMAT and CAVMAT plans was performed using Delta4 and SRS MapCHECK with 3% / 1mm, 2% / 1mm, 1% / 1mm criteria to assess the agreement during patient specific quality assurance.
Result: For Portal Dosimetry QA of IMRT brain and SBRT/HIGRT using a 3%/1mm criteria, the TG-218 action limit was 99.68, and 90.14, respectively; with 3.68% and 3.68% of cases failing the universal 90% criteria. For Delta4 QA of single target SRS, multiple target SRS, and SBRT IMRT with FFF using a 3%/1mm criteria, the TG-218 action limit was 93.64, 97.12, and 92.01, respectively; with 0%, 0%, and 0% of cases failing the universal 90% criteria. For Delta4 QA of SBRT VMAT using a 4%/1mm criteria, the TG-218 action limit was 94.47, with 100% passing. For ArcCHECK QA of single target and multiple target SRS VMAT using a 3%/2mm criteria, the TG-218 action limit was 98.06 and 96.59 respectively, with 100% passing. For SRS MapCHECK QA of multiple target SRS VMAT cases using 3%1mm criteria, the TG-218 action limit was 99.24 with 100% passing.
The average increase in V6Gy, V12Gy, V16Gy due to treatment delivery errors as quantified using the trajectory logfile was 0.94 ± 1.43, 0.90 ± 1.38%, and 1.23 ± 1.54% respectively for VMAT, and 0.035 ± 0.14%, 0.14 ± 0.18%, and 0.28 ± 0.24% for CAVMAT. The average change to target maximum, average, and minimum dose due to delivery errors was 0.53 ± 0.46%, 0.52 ± 0.46%, and 0.53 ± 0.56%, for VMAT, and 0.16 0.18%, 0.11 0.08%, and 0.03 0.24% for CAVMAT. There was no significant difference in magnitude of MLC discrepancies during delivery for VMAT and CAVMAT. For Gamma analysis with strict 1% / 1mm criteria, the average passing rate of VMAT gamma analysis is 94.53 ± 4.42%, while that of CAVMAT is 99.28 ± 1.74%.
Conclusion: For most QA devices, spatial tolerance of pre-treatment QA for SRS/SBRT can be tightened to 1mm while still maintaining an in-control QA process. The gamma criteria to 3%/1mm for all SRS cases and SBRT with IMRT and transitioning to a 4%1mm criteria for SBRT with VMAT have a spatial tolerance that is appropriate for the radiotherapy technique while not resulting in an excessive false positive failure rate. The CAVMAT treatment planning technique resulted in superior gamma analysis passing rate for each gamma analysis criteria.
Item Open Access Assessing the feasibility of using deformable registration for on-board multi-modality based target localization in radiation therapy(2018) Ren, GePurpose:
Cone beam computed tomography (CBCT) is typically used for on-board target localization in radiation therapy. However, CBCT has poor soft tissue contrast, which makes it extremely challenging to localize tumors in soft tissues such as liver, prostate and breast cancers. This study explores the feasibility of using deformable image registration (DIR) to generate on-board multi-modality images to improve the soft tissue contrast for target localization in radiation therapy.
Methods:
Patient CT or MR images are acquired during the simulation stage and are used as the prior images. CBCT images are acquired on-board for clinical target localization. B-spline based deformable registration is used to register MR or CT images with CBCT images to generate synthetic on-board MR/CT images, which are used for on-board target localization. Liver, prostate, and breast patient data were used in the study to investigate the feasibility of the method. The evaluation includes three aims: (1). Evaluate whether the registration and margin design in clinical practice is sufficient to ensure the coverage of the on-board tumor volume: the synthetic on-board MR/CT images are used to verify the target coverage based on the shifts determined by CT-CBCT registration in clinical practice; (2). Evaluate the potential for margin optimization based on the synthetic multi-modality imaging technique: shifts are determined by rigid registration between planning CT and synthetic on-board MR/CT, and the replacing PTV margin is determined to ensure coverage of the deformed tumor volume. (3). Evaluate the potential tolerant margin for DIR uncertainty based on the deformed tumor contour in planning CT images: shifts are determined by rigid registration between planning CT and the synthetic on-board MR/CT, and the tolerant margin is determined to cover the expanded deformed tumor volume in PTV.
Results:
In the process of DIR, using CT images as DIR prior images has better alignment than using MR as the prior images. The evaluation showed: (1). For the liver cases, the coverage of 6 in 8 cases is above 90%. For the breast cases, the coverage of 6 in 7 cases is above 90%. For the prostate cases, the coverage of all cases is above 94%. Most of the tumor volume defined by the on-board synthetic images were covered by the PTV based on the shifts applied in clinical practice. The 3 under-dosed cases are correlated with long,interfraction deviation treatment fraction, small volume, and zero-PTV margin design. (2). For 6 of the liver cases, 5 of the prostate cases, and all the breast cases, the synthetic images allowed the reduction of PTV margin, which is up to 6mm, 4mm, and 1.5 mm, respectively. For the cases with reduced optimized margin, the dose to the OAR and normal tissue can be spared based on the optimized margin while for the cases with increased optimized margin the increased dose is not significant. (3). For cases with reduced margin, the benefit margin for DIR uncertainty is available which are 2-4 mm, 1-5mm, and 2-3 mm for liver, prostate, and breast cases respectively.
Conclusion:
Our studies demonstrated the feasibilities of using on-board synthetic multi-modality imaging to improve the soft tissue contrast for target localization in low contrast regions. This new technique holds great promises to optimize the PTV margin and improve the treatment accuracy.
Item Open Access Beam Optimization for Whole Breast Radiation Therapy Planning(2018) Wang, WentaoPurpose: To develop an automated program that can generate the optimal beams for whole breast radiation therapy (WBRT).
Methods and Materials: A total of twenty patients receiving WBRT were included in this study. The computed tomography (CT) simulation images and structures of all 20 patients were used to develop and validate the program. All patients had the breast planning target volume (PTV) contour drawn by physicians and radio-opaque catheters placed on the skin during CT simulation. First, an initial beam was calculated based on the CT images, the radio-opaque catheters, and the breast PTV contour. The beam includes five main parameters: the gantry angles, the isocenter location, the field size, the collimator angles, and the initial multi-leaf collimator (MLC) shape.
To optimize the beam parameters, a geometry-based objective function was constructed to optimize target coverage and organ-at-risk (OAR) sparing. The objective function is the weighted sum of the square of the relative volumes of the PTV outside the field and the ipsilateral lung inside the field. Due to the curvature of the chest wall, a portion of the ipsilateral lung will be included in the irradiated volume. The balance between PTV coverage and OAR sparing is embodied by the relative weight of the lung volume in the objective function, which was trained and validated from the clinical plans of the twenty patients. Two different optimization schemes were developed to minimize the objective function: the exhaustive search and the local search. The search was conducted in a 2-dimensional grid with the gantry angle (1° increments) and the isocenter location (1 mm increments) as two axes and the initial beam as the origin point. For the exhaustive search, the ranges of the gantry angle and the isocenter location are ±12° and ±21 mm. The local search does not require a search range. The beam with the minimal objective function value in the grid is considered optimal. The optimal beam was transferred to an in-house automatic fluence optimization program developed specifically for WBRT. The automatic plans were compared with the manually generated clinical plans for target coverage, dose conformity and homogeneity, and OAR dose.
Results: The calculation time of the beam optimization was under one minute for all cases. The local search (~15 s) took less time than the exhaustive search (~45 s), and the two methods produced the same result for the same patient. The automatic plans have overall comparable plan quality to the clinical plans, which usually take 1 to 4 hours to make. Generally, the PTV coverage is improved while the dose to the ipsilateral lung and the heart is similar. The breast PTV Eval V95% of all cases are above 95%, and the mean V95% (97.7%) is increased compared with the clinical plans (96.8%). The ipsilateral lung V16Gy is reduced for 14 out of 20 cases, and the mean V16Gy is decreased in the automatic plans (12.6% vs. 13.6%). The average heart mean dose is slightly increased in the automatic plans (2.06% vs. 1.99%).
Conclusion: Optimal beams for WBRT can be automatically generated in one minute given the patient’s simulation CT images and structures. The automated beam setup program offers a valuable tool for WBRT planning, as it provides clinically relevant solutions based on previous clinical practice as well as patient specific anatomy.
Item Open Access Deep Learning for Automatic Real-time Pulmonary Nodule Detection and Quantitative Analysis(2019) Liu, ChenyangPurpose: To develop a novel computer-aided diagnosis (CAD) pulmonary nodule detection system that can not only perform real-time detection but also characterize quantitative nodule information based on deep learning methods.
Method: We constructed a convolutional neural network (CNN) for automated pulmonary nodule detection and characterization. Nodule detection was accomplished by customizing a detection algorithm (YOLO v3), which comprised of a feature extractor and a bounding box generator. The feature extractor had 19 convolutional layers with 7 residual shortcut connections to extract features on input images at three different down-sampling scales (i.e. 4, 8, and 16). The bounding box generator had 7 convolutional layers to determine the location and size of each detected nodule. A python-based characterization system was then developed to characterize size, diameter, and central coordinates of each detected nodule within the generated bounding box. This characterization system applied a non-maximum suppression algorithm to exclude nodules below true positive probability threshold. The system was trained and validated using ten-fold cross-validation with 300 CT scans from XCAT simulation and 888 patient CT scans from LIDC–IDRI public dataset, separately. System performance was evaluated using Free-Response Receiver Operating Characteristic (FROC) analysis, competition performance metric (CPM) score, as well as precision analysis of central coordinates and diameters.
Result: The developed CAD system achieved CPM scores of 0.99 in the simulation image study and 0.873 in the public database study. The average performance time per image was less than 0.1 second. Compared with ground truth data, the detection precision in diameter were 0.26 mm using simulated images and 1.05 mm using public database, while the precision in central coordinate were 0.76 mm and 1.44 mm, respectively.
Conclusion: Preliminary evaluation showed that our proposed CAD system using deep learning methods was robust and achieved real-time nodule detection with high accuracy and characterization with high precision.
Item Open Access Deep Learning Segmentation in Pancreatic Ductal Adenocarcinoma Imaging(2024) Zhang, HaoranPurpose: Accurately quantifying the extent vessel coverage in the pancreas is essential for determining the feasibility of surgery. The aim of this study is to train a segmentation model specialized for Pancreatic Ductal Adenocarcinoma (PDAC) imaging, focusing on delineating pancreas, tumor, arteries (celiac artery, superior mesenteric artery, common hepatic artery), and veins (portal vein, superior mesenteric vein) using an improved Attention Unet CNN approach.Methods: Data from 100 PDAC patients treated at the Ruijin Hospital between 2020 and 2022 were utilized. Using Synapse 3D software, masks of the tumor, arteries (including the celiac artery, superior mesenteric artery, and hepatic artery), and veins (including the superior mesenteric vein and portal vein) were generated semi-automatically and reviewed by radiologists. Standard image processing techniques, including adjustment of window level to 60 and width to 350 and histogram equalization were subsequently applied. Two types of CNN-based Attention Unet segmentation models were developed: (1) Unified Unet Model that segments all four components simultaneously, and (2) four Individual Unet Models that segments pancreas, tumor, veins, and arteries separately. The train-validation-test data assignment was set to 7:2:1. The segmentation efficacy was assessed using Dice similarity coefficient, with the Adam optimizer utilized for optimization. Results: The individual segmentation models achieve notable performance: pancreas (Accuracy: 0.84, IoU: 0.81, Dice: 0.76), tumor (Accuracy: 0.78, IoU: 0.77, Dice: 0.68), vein (Accuracy: 0.88, IoU: 0.86, Dice: 0.80), and artery (Accuracy: 0.91, IoU: 0.93, Dice: 0.93). However, the unified model demonstrates inferior performance with accuracy, IoU, and Dice coefficient scores of 0.61, 0.50, and 0.45, respectively. Conclusion: Accurate segmentation models have been developed for pancreas, pancreatic tumors, arteries, and veins in PDAC patients. This will enable the efficient quantification of vessel coverage in the pancreas, thereby enhancing the decision-making process regarding the feasibility of surgery for PDAC patients. The findings also demonstrate that Individual Models outperform the Unified Model in segmentation accuracy, highlighting the importance of tailored segmentation strategies for different anatomical structures in PDAC imaging.
Item Embargo Deep Learning-based Brain Image Segmentation on Turbo Spin Echo MRI(2024) Zhang, TianyiPurpose: Currently, the Magnetization Prepared Rapid Gradient Echo (MPRAGE) Magnetic Resonance Imaging (MRI) sequence is frequently used for brain tissue segmentation in the clinic due to its high image contrast. However, one of the limitations of the MPRAGE sequence lies in its susceptibility to metal artifacts, while the Turbo Spin Echo (TSE) sequences, can resist metal artifacts. Previous studies have shown that for patients with metal implants, metal-artifact-reduced MPRAGE images can be generated from TSE images. Conventional brain segmentation methods on MPRAGE images, such as FreeSurfer, are time-consuming. Therefore, the purpose of this study was to investigate a fast brain segmentation method via deep learning-based frameworks for patients with metal implants, using TSE images as input.Materials and Methods: A dataset consisting of 369 patients in total was used. Each patient contained 160 two-dimensional slices of T1-weighted (T1WI), T2-weighted (T2WI), and PD-weighted (PDWI) TSE brain MR images, respectively. The matrix size of the original images was 240 × 240. Two types of MPRAGE as intermediate steps were synthesized from T1WI, T2WI, and PDWI using mathematical calculations or Conditional Generative Adversarial Network (cGAN) algorithms. FreeSurfer software was used to generate brain segmentations on the MPRAGE, which were considered as the ground truth for deep-learning network training and eventual evaluation. Two research aims were investigated. Aim 1 was to utilize three-channel TSE images (T1WI, T2WI, and PDWI) to first mathematically synthesize MPRAGE images, and then perform segmentation via deep learning-based models. Aim 2 was to use single-channel TSE images as input directly or indirectly to achieve brain segmentation using deep learning-based models. Both UNet and UNet++ models were examined. The Dice coefficient was used to evaluate the performance of the above-mentioned segmentation aims. Results: For Aim 1, the Dice coefficient between the ground truth and the cortex segmentations generated by the UNet++ network using three-channel TSE images as original input and mathematically synthesized MPRAGE as direct input was 0.919 ± 0.03. For Aim 2, the Dice coefficient between the ground truth and the cortex segmentations generated by the UNet network using single-channel TSE images directly as input was 0.602 ± 0.06. The Dice coefficient between the ground truth and the cortex segmentations generated by the single-channel TSE images as original input and cGAN-synthesized MPRAGE as direct input using the UNet++ network was 0.766 ± 0.07. Conclusion: Two aims using three-channel or single-channel TSE images as original input and brain segmentation as output were investigated in this study. Three-channel TSE images as original input, and mathematically synthesized MPRAGE as direct input to the UNet++ network showed superior results. Single-channel TSE images as original input and cGAN-synthesized MPRAGE as direct input to the UNet++ network showed relatively lower performance. Further research is warranted to improve the performance of single-channel TSE-based deep-learning segmentation methods. Keywords: UNet++, MRI, Brain Image, Segmentation, TSE, MPRAGE
Item Open Access Development and Evaluation of a Bi-polar Gated Respiratory Motion Management Strategy for Lung SBRT(2021) Li, ZhenBackgroundStereotactic body radiation therapy (SBRT) is a noninvasive alternative treatment for patients who cannot accept surgery. It delivers higher ablative total radiobiological doses to the tumor in fewer fractions. The dose can be as high as 20Gy per fraction (compared to 1.8 to 2 Gy in conventional radiotherapy treatment) and the number of fractions are usually less than 5 (compared to 30 in conventional treatment). Thus, it requires high conformal dose distribution and minimal exposure of surrounding healthy tissues for the patients. The major challenges of SBRT for lung cancer include respiratory motion, positioning uncertainty and intra-fraction stability. Breath hold is usually not suitable for lung cancer patients because they cannot hold the breath for a long time (typically 20 seconds or more) for treatment. Current treatment strategies for Lung SBRT include free breathing (FB), gating at the end of exhale (GE), gating at the end of inhale (GI), and real-time tracking (RT), which are illustrated in the figure below (Fig.a). Most lung SBRT patients are treated in free breathing. It delivers dose to all the tumor motion trajectory, thus requires field size larger enough to cover all the possible tumor locations, creating extra dose to normal tissue. This results in lesion size limitation, so FB is most widely used in small tumors or tumors with less motion, such as those located in the upper lobes of the Lung. The advantages of FB are its low technical requirements and high treatment efficiency (100% duty cycle). For gating technology, the radiation is limited to several specific respiratory phases at the end of exhale or inhale in which the target motions are reduced. It monitors motion and reduce treatment volume, which offers the dosimetric advantage of lower doses to organs at risk. The duty cycle of gating depends on the threshold and is usually less than 30%. Gated treatment can be executed in two strategies: GE and GI. These two strategies each has their own pros and cons. For GE, the tumor position is typically more reproducible than GI. However, GI has larger lung volume due to inspiration and this can lead to lower lung toxicity. In real-time tracking, the beam moves with the tumor movement and delivers the dose to the exact tumor position. It is the most effective strategy because it has theoretically zero motion margin. Another advantage is high treatment efficiency with the beam always on during treatment. The drawback of RT is that it is highly technical demanding and there may be some time delay in tumor detection, decision making and taking actions. Therefore, it is more complex than other strategies and could result in additional potential errors. In addition, the dose delivery accuracy of real-time tracking is limited by the 4DCT resolution and this will be introduced in the following section.
Purpose To reduce the motion margin in dose delivery, we developed a novel bipolar (BP) strategy that requires the patient to hold the breath at the end of the inhale and exhale under audio/video coaching. By only delivering the dose to the tumor during breath-holding, the dose is given to the exact tumor position. This reduces normal tissue toxicities to a great extent. Moreover, the treatment duty cycle is higher than conventional gating technology. In this study, we want to evaluate the feasibility of BP breathing pattern and also compare BP with other strategies geometrically and dosimetricly. Methods BP Feasibility Analysis: Feasibility analysis is conducted to see if the bi-polar breathing pattern is sustainable and comfortable for patients to breathe in a long time. 8 volunteers are included in this study to breathe following the FB/RT, GE, GI and BP breathing patterns under audio coaching. The respiratory signals acquisition time for each pattern is more than three minutes. A custom MatLab program is developed for data analysis. The period repeatability, breath-hold repeatability and treatment efficiency are calculated and compared for each strategy. Geometric Evaluation: 10 previously treated lung SBRT patients with 4DCT were selected retrospectively, each having tumor motion ≥ 1cm. The tumor size at end-exhale (EE) ranges from 0.1 to 22.7 and 80% cases larger than 1cc. 60% tumors located at the lower lobe of the lung. Lung volume and tumor position were used to determine the end-exhale (EE) and end-inhale (EI) phases. The GTV was contoured at each 4DCT phase to determine the ITV for each strategy. PTV is formed by 1mm expansion from ITV. The lung volume, ITV and PTV in BP were compared with FB, GE, GI and RT. All the values are normalized to GI to include all the patients for comparison. Dosimetric Evaluation: IMRT and VMAT plans were generated for each patient with a prescription dose of 60 Gy in 5 fractions. All the plans are completed in the Varian Eclipse System (Version 15.5). The energy we used is 6MV and the calculation algorithm is AAA. 100% dose is normalized to 95% volume. All the plans should meet the RTOG 0813 protocol. IMRT uses 7-9 beams and VMAT uses 1-2 arcs. OARs includes lungs, spinal cord, esophagus, trachea, heart etc., were contoured. For BP and RT, a custom MatLab program was used to summate the plans and calculate the DVHs. Parameters include V5Gy, V13Gy, V20Gy and MLD (mean lung dose) were compared for each strategy in both VMAT and IMRT plans. Tumor Motion Modeling: The purpose of this section is to prove the observed volume in 4DCT is larger than real tumor volume due to tumor motion and the limited number of phases in 4DCT. Both phantom and patient study are included in tumor motion modeling to verify our assumption. QUASAR phantom with a white ball inside (radius: 1.5cm) was used for phantom data acquisition. 5 breathing patterns using same motion amplitude was acquired. The BPM (breath per minute) are 15, 20, 25, 30,33, respectively. 10 previously treated lung SBRT patients with 4DCT were selected for patient study. These patient data are the same as patient data in geometric evaluation section. We developed a MatLab program to calculate the theoretical volume (simulated volume) in each 4DCT phase. By comparing the simulated volume with the observed volume, we want to verify that the observed tumor volume is larger than the real tumor volume in each 4DCT phase and it is a function of real tumor volume and tumor motion.
ResultsFeasibility Analysis: BP breathing pattern is found to be comfortable and sustainable over 3 minutes. This may be longer if we test for longer time. The period repeatability and breath-hold repeatability are at 1.00±0.03 and 1.00±0.04. It is higher than GE(with breath-hold) and GI (with breath-hold), indicating a better repeatability for BP. Treatment efficiency of BP can be more than 65% under audio coaching. It may be improved with the video coaching, longer period of breath holding and patient training. Geometric Evaluation: Using GI as reference, ITV in FB is the largest among all 5 strategies and it is significantly larger than BP. That’s because FB delivers dose to the whole tumor motion trajectory thus creating a large tumor motion margin. The ITVs in RT and BP are similar and smaller. They are approximately one third of FB. The ITV in FB is about twice of the ITVs in GE and GI. Generally, PTV shows a similar trend with ITV. FB is significantly larger than other strategies and it is approximately 2.5 times of RT and BP. The PTV of GE and GI are similar and they are about 56% of the FB. BP is a little bit smaller than RT because the fast-moving tumor in limited phases. Comparing with ITV and PTV, they basically follow the same trend. However, the difference between BP and FB are narrowed due to 1mm expansion from ITV to PTV. In PTV, BP is 58% less than FB, and in ITV, BP is 67% less than FB. Dosimetric Evaluation: In IMRT, all the dose are normalized to GI. FB is the highest for V5Gy, V13Gy, V20Gy and MLD in all strategies and BP is significantly smaller than FB. The reason for that is the largest PTV for FB and smallest PTV for BP. V5Gy, V13Gy, V20Gy and MLD in RT and BP are similar and smaller than GE, GI. They are approximately 20%-30% lower than FB. Although GE and GI have similar PTVs, the dose in GI for V5Gy, V13Gy, V20Gy and MLD are much smaller than GE due to lager lung volume in GI. In VMAT, the evaluation parameters are the same as in IMRT. They basically show similar trend with IMRT. The dose for all parameters in FB is the largest and BP is also significantly smaller than FB. RT and BP is similar and smaller and they are approximately 10%-20% lower than FB for all the parameters. These values are smaller than IMRT. Overall, the improvement from FB to BP is slightly larger in IMRT than VMAT.
Tumor Motion Modeling:In phantom data analysis, for all the cases, the simulated volume achieves minimum at EE and EI phases due to minimum motion, and it increases with larger motion. Observed volumes agree with simulated volume at most phases. In patient data analysis, the agreement is not as good as phantom. The simulated volume achieves its minimum at EE and EI phases and the tumor volume is larger in other phases. Although the observed volumes do not perfectly agree with the simulated volume for most patient cases, they basically follow the same trend. The possible reasons can be tumor location (connecting to diaphragm or vessels) and patients’ irregular respiratory repeatability. More patient data with clear and isolated margin should be included in the future. Conclusion The respiratory experiment demonstrates that the bi-polar breathing pattern is feasible for lung SBRT. It can sustain a long treatment time with a high duty cycle. Moreover, compared to the breathing pattern of GE and GI, BP is more regular, comfortable and thus more sustainable than other breathing patterns. The tumor volume in each 4DCT phase can be inaccurate due to tumor motion and limited resolution in 4DCT. It is actually a function of real tumor volume and tumor motion. Thus, the treatment volume in RT overestimates the actual tumor volume since it used the observed volume as GTV. For the tumors with larger motion (≥1cm) in this research, BP has a significantly smaller ITV in geometric evaluation; therefore, a smaller treatment volume (PTV) compared to FB, GE and GI. The dosimetric evaluation of IMRT and VMAT shows lower V5Gy, V13Gy, V20Gy, and MLD in BP, especially when comparing to FB. This will lead to lower lung dose in treatment. RT has similar geometric and dosimetric results with BP. However, RT is more complicated than BP in implementation and dose delivering.
Item Open Access Development and Testing of An Automatic Lung IMRT Planning Algorithm(2016) Zhu, WeiKnowledge-based radiation treatment is an emerging concept in radiotherapy. It
mainly refers to the technique that can guide or automate treatment planning in
clinic by learning from prior knowledge. Dierent models are developed to realize
it, one of which is proposed by Yuan et al. at Duke for lung IMRT planning. This
model can automatically determine both beam conguration and optimization ob-
jectives with non-coplanar beams based on patient-specic anatomical information.
Although plans automatically generated by this model demonstrate equivalent or
better dosimetric quality compared to clinical approved plans, its validity and gener-
ality are limited due to the empirical assignment to a coecient called angle spread
constraint dened in the beam eciency index used for beam ranking. To eliminate
these limitations, a systematic study on this coecient is needed to acquire evidences
for its optimal value.
To achieve this purpose, eleven lung cancer patients with complex tumor shape
with non-coplanar beams adopted in clinical approved plans were retrospectively
studied in the frame of the automatic lung IMRT treatment algorithm. The primary
and boost plans used in three patients were treated as dierent cases due to the
dierent target size and shape. A total of 14 lung cases, thus, were re-planned using
the knowledge-based automatic lung IMRT planning algorithm by varying angle
spread constraint from 0 to 1 with increment of 0.2. A modied beam angle eciency
index used for navigate the beam selection was adopted. Great eorts were made to assure the quality of plans associated to every angle spread constraint as good
as possible. Important dosimetric parameters for PTV and OARs, quantitatively
re
ecting the plan quality, were extracted from the DVHs and analyzed as a function
of angle spread constraint for each case. Comparisons of these parameters between
clinical plans and model-based plans were evaluated by two-sampled Students t-tests,
and regression analysis on a composite index built on the percentage errors between
dosimetric parameters in the model-based plans and those in the clinical plans as a
function of angle spread constraint was performed.
Results show that model-based plans generally have equivalent or better quality
than clinical approved plans, qualitatively and quantitatively. All dosimetric param-
eters except those for lungs in the automatically generated plans are statistically
better or comparable to those in the clinical plans. On average, more than 15% re-
duction on conformity index and homogeneity index for PTV and V40, V60 for heart
while an 8% and 3% increase on V5, V20 for lungs, respectively, are observed. The
intra-plan comparison among model-based plans demonstrates that plan quality does
not change much with angle spread constraint larger than 0.4. Further examination
on the variation curve of the composite index as a function of angle spread constraint
shows that 0.6 is the optimal value that can result in statistically the best achievable
plans.
Item Open Access Development of a Voxel-Based RadiomicsCalculation Platform for Medical Image Analysis(2020) Yang, ZhenyuPurpose: To develop a novel voxel-based radiomics extraction technique, and to investigate the potential association between spatially-encoded radiomics features of the lungs and pulmonary function.
Methods: We developed a voxel-based radiomics feature extraction platform to generate radiomics filtered images. Specifically, for each voxel in the image, 62 radiomics features were calculated in a rotationally-invariant 3D neighbourhood to capture spatially-encoded information. In general, such an approach results in an image tensor object, i.e., each voxel in the original image is represented by a 62-dimensional radiomics feature vector. Two digital phantoms are then designed to validate the technique's ability to quantify regional image information. To test the technique as a potential pulmonary biomarker, we generated radiomics filtered images for 25 lung CT image and are subsequently evaluated against corresponding Galligas PET images, as the ground truth for pulmonary function, using voxel-wise Spearman correlation (r). The Canonical Correlation Analysis (CCA)-based feature fusion method is also implemented to enhance such a correlation. Finally, the Spearman distributions were compared with 37 individual CT ventilation image (CTVI) algorithms to assess the overall performance relative to conventional CT-based techniques.
Results: Several radiomics filtered images were identified to be correlated with Galligas PET lung imaging. The most robust association was found to be the Run Length Encoding feature, Run-Length Non-uniformity (0.21
Conclusions: This preliminary study indicates that spatially-encoded lung texture and lung density are potentially associated with pulmonary function as measured via Galligas PET ventilation images. Collectively, low density, heterogeneous coarse lung texture was often associated with lower Galligas radiotracer amounts.
Item Open Access Dosimetric Evaluation of Metal Artefact Reduction using Metal Artefact Reduction (MAR) Algorithm and Dual-energy Computed Tomography (CT) Method(2016) Laguda, Edcer Jerecho Dela CruzPurpose: Computed Tomography (CT) is one of the standard diagnostic imaging modalities for the evaluation of a patient’s medical condition. In comparison to other imaging modalities such as Magnetic Resonance Imaging (MRI), CT is a fast acquisition imaging device with higher spatial resolution and higher contrast-to-noise ratio (CNR) for bony structures. CT images are presented through a gray scale of independent values in Hounsfield units (HU). High HU-valued materials represent higher density. High density materials, such as metal, tend to erroneously increase the HU values around it due to reconstruction software limitations. This problem of increased HU values due to metal presence is referred to as metal artefacts. Hip prostheses, dental fillings, aneurysm clips, and spinal clips are a few examples of metal objects that are of clinical relevance. These implants create artefacts such as beam hardening and photon starvation that distort CT images and degrade image quality. This is of great significance because the distortions may cause improper evaluation of images and inaccurate dose calculation in the treatment planning system. Different algorithms are being developed to reduce these artefacts for better image quality for both diagnostic and therapeutic purposes. However, very limited information is available about the effect of artefact correction on dose calculation accuracy. This research study evaluates the dosimetric effect of metal artefact reduction algorithms on severe artefacts on CT images. This study uses Gemstone Spectral Imaging (GSI)-based MAR algorithm, projection-based Metal Artefact Reduction (MAR) algorithm, and the Dual-Energy method.
Materials and Methods: The Gemstone Spectral Imaging (GSI)-based and SMART Metal Artefact Reduction (MAR) algorithms are metal artefact reduction protocols embedded in two different CT scanner models by General Electric (GE), and the Dual-Energy Imaging Method was developed at Duke University. All three approaches were applied in this research for dosimetric evaluation on CT images with severe metal artefacts. The first part of the research used a water phantom with four iodine syringes. Two sets of plans, multi-arc plans and single-arc plans, using the Volumetric Modulated Arc therapy (VMAT) technique were designed to avoid or minimize influences from high-density objects. The second part of the research used projection-based MAR Algorithm and the Dual-Energy Method. Calculated Doses (Mean, Minimum, and Maximum Doses) to the planning treatment volume (PTV) were compared and homogeneity index (HI) calculated.
Results: (1) Without the GSI-based MAR application, a percent error between mean dose and the absolute dose ranging from 3.4-5.7% per fraction was observed. In contrast, the error was decreased to a range of 0.09-2.3% per fraction with the GSI-based MAR algorithm. There was a percent difference ranging from 1.7-4.2% per fraction between with and without using the GSI-based MAR algorithm. (2) A range of 0.1-3.2% difference was observed for the maximum dose values, 1.5-10.4% for minimum dose difference, and 1.4-1.7% difference on the mean doses. Homogeneity indexes (HI) ranging from 0.068-0.065 for dual-energy method and 0.063-0.141 with projection-based MAR algorithm were also calculated.
Conclusion: (1) Percent error without using the GSI-based MAR algorithm may deviate as high as 5.7%. This error invalidates the goal of Radiation Therapy to provide a more precise treatment. Thus, GSI-based MAR algorithm was desirable due to its better dose calculation accuracy. (2) Based on direct numerical observation, there was no apparent deviation between the mean doses of different techniques but deviation was evident on the maximum and minimum doses. The HI for the dual-energy method almost achieved the desirable null values. In conclusion, the Dual-Energy method gave better dose calculation accuracy to the planning treatment volume (PTV) for images with metal artefacts than with or without GE MAR Algorithm.
Item Open Access Dynamic shimming of the human brain with a 32-channel integrated parallel reception, excitation, and shimming (iPRES) head coil array(2018) Zhang, HuiminIntegrated parallel reception, excitation, and shimming (iPRES) is a novel MRI coil design, in which radiofrequency (RF) currents and direct currents (DC) flow in the same coil elements to perform MR image acquisition and localized magnetic field shimming, respectively, with a single coil array. The purpose of this study was to implement dynamic shimming with a 32-channel iPRES head coil array, by dynamically updating the DC currents to shim individual slices within a single scan. Dynamic shimming is more effective than global static shimming, in which the same currents are used to shim the whole brain, and more efficient than slice-optimized static shimming, in which different currents are used to shim different slices, but in separate scans.
To implement dynamic shimming, a Python script was written to send new DC current amplitudes and polarities to a DC power supply and a switch box connected to the iPRES head coil array, respectively, for each slice acquisition. This current-update process was optimized by performing timing measurements with an oscilloscope and by modifying the Python script to ensure that the DC currents to shim the ith slice were updated as efficiently as possible after the data acquisition of the (i-1)th slice and before the excitation of the ith slice.
Magnetic field maps were acquired in a phantom and in a human brain with either dynamic shimming or slice-optimized static shimming, and the root-mean-square-error was calculated to evaluate the shimming performance of the dynamic shimming relative to slice-optimized static shimming. The results show that dynamic shimming with the iPRES head coil array was successfully implemented and that it was as effective as, but much more efficient than, slice-optimized static shimming.
Item Embargo Efficient Synthesizing High Quality Digital Cortex Phantoms through a Novel Combination of GAN and CNN Methods(2024) Pan, JiongliPurpose: Imaging data scarcity is an intrinsic challenge in medical research partly due to resource limitations and privacy concerns. Digital human phantom is one approach to mitigate such challenge. The aim of this study is to propose an innovative method combining Generative Adversarial Network (GAN) and Convolutional Neural Network (CNN) to virtually synthesize large scale and high-quality brain cortex digital phantoms.Method: High resolution brain MRIs (MP-RAGE) of 369 patients were used to generate brain segmentations using FreeSurfer software. Total of 14,200 segmented 2D cortex phantoms were extracted and served as ground truth. A GAN was developed to learn and synthesize new cortex phantoms. 28,955 phantoms were generated and manually classified into ‘real’ and ‘fake’ groups, where 14,364 were labeled as ‘real’ while rest were labeled as ‘fake’ due to mal/poor morphology and/or noise. To facilitate automatic phantom quality classification, a CNN was designed and trained on the total of 43,155 phantoms, where the training and testing split was 7:3. Receiver Operating Characteristic Curve (ROC) analysis and Area Under the Curve (AUC) evaluation were performed on the CNN. The GAN and CNN together constitute the automatic cortex digital phantom generation network. To visually evaluate the realism of the model-generated phantoms, two experiments were conducted: the first involved binary classification of real and generated phantoms, and the second added a ‘Cannot Decide’ option to the previous task. Eight volunteers were recruited for these experiments. In addition to visual assessments, the Fréchet Inception Distance (FID) was employed as a quantitative metric to evaluate the generated phantoms. Results: The CNN-based filtering network attained an accuracy of 97% and an AUC of 0.99. This integrated GAN generation and CNN classification network generated over 130,000 high-quality phantoms in less than one hour on a PC (NVIDIA GeForce GTX 1060), with only 6 phantoms contained minor noise on the phantom border. The morphological quality of the final phantoms received positive approval from visual evaluation experiments, demonstrating that the ability to distinguish synthetic from real phantoms in paired comparisons was about 50%. The FID score of 9.61 suggests a high level of diversity in the synthetic phantoms. Conclusion: This study demonstrated a novel and efficient method for generating a large-scale synthetic medical digital phantom by integrating two deep-learning models, GAN and CNN. The synthesized digital phantoms not only showed realistic morphology but also maintains commendable data diversity. Additionally, this approach offers a solution to the time-consuming and subjective nature of manually assessing the phantom quality generated by GAN networks. The digital phantoms synthesized have potential to be utilized in various phantom research areas.
Item Unknown Iterative Reconstruction of SPECT Brain with Priors Based on MRI T1 and T2 Images(2017) Gu, QinglongPurpose: Although brain Single Photon Emission Computed Tomography (SPECT) exam is a low cost, widely used functional and molecular brain imaging application, it also has poor spatial resolution (64 x 64 or 128 x 128, pixel sizes about 3 - 6 mm) and a noisy signal. As a result, the SPECT brain images may not be quantitatively accurate for radiotracer uptake, mainly gray matter (GM) and white matter (WM). Many studies have considered improving SPECT quantification by incorporating Magnetic Resonance Imaging (MRI) images into SPECT images. MRI has much higher spatial resolution (192x192 or 256 x 256, pixel sizes 1 to 1.5 mm), which is useful in correcting partial-volume degradation of SPECT quantification. MRI also provides broader image contrast options with many different types of MRI sequences, typically in T1 weighted (T1WI) and T2 weighted (T2WI) sequences. In most previous studies into the use of MRI or CT images to generate the anatomical priors for SPECT/ Positron Emission Tomography (PET) image reconstruction, only a single MRI sequence has been considered. Few studies have investigated the effects of different MRI sequence on the anatomical prior and the resulting SPECT/PET based on the different MRI sequences. In the present study, we evaluate the SPECT brain images at the midbrain level, with the anatomical priors based on the MRI T1WI gradient echo (GE) images and T2WI fast spin echo (FSE) images.
Materials and methods: Source brain images were downloaded from BrainWeb for SPECT image simulation. These included fuzzy gray matter and white matter models for digital radiotracer phantom creation, MRI T1WI and T2WI images for use in SPECT image-reconstruction anatomical priors. The images were selected at the midbrain level and converted to 32 bit to be used in SPECT-MAP. In SPECT-MAP, a ground truth radiotracer phantom was generated for Tc99m-ECD brain perfusion studies. Based on the phantom, SPECT projection data were simulated. These simulations modeled noise and spatial resolution. SPECT images were then reconstructed by maximum a posteriori (MAP). The prior probability distributions were generated from either gradient-echo T1 or fast-spin-echo T2 MRI images. The MAP objective function was optimized using an iterative coordinate descent (ICD) algorithm. SPECT images were also reconstructed by ordered subsets expectation maximization (OSEM). Reconstructed images were compared to the true phantom radiotracer distribution by visual inspection profiles, root mean square error (RMSE), and gray matter to white matter contrast to deviation ratio (CDR).
Results: After 17 iterations, the RMSE for method T1 MAP, T2 MAP and OSEM was 1266.4, 1752.6 and 3231.9. The CDR for method T1 MAP, T2 MAP and OSEM was 10.1, 7.8 and 2.8, which in the digital radiotracer phantom was 25.1. Relative to the T2-based prior, utilizing the T1-based prior for SPECT image reconstruction improved RMSE and CDR by 18% and 29% respectively. Relative to the best iterations for OSEM, the T1-based prior improved RMSE by 43% and CDR by a factor of 2.6. Visually, the SPECT image reconstructed with the T1-based prior was closest to the true phantom distribution, notably capturing certain structures that were not well reconstructed using the T2 image. Both MAP images were superior to OSEM visually and by RMSE and CDR.
Conclusion: The quality of SPECT images reconstructed utilizing MRI images depends substantially on the MRI sequence utilized. For this study, gradient-echo T1 MRI provided more accurate SPECT image reconstruction than fast-spin-echo T2 MRI. Both MRI sequences resulted in better RMSE and CDR than OSEM without use of MRI. The CSF signal distorted MRI boundaries relative to radiotracer boundaries, particularly for MRI T2 sequences. A T2 FLAIR-like images improved boundary alignment and SPECT reconstructed image as compared to T2 MRI images when they were used in anatomical priors for SPECT image reconstruction.
Item Unknown LOW DOSE CT ENHANCEMENT USING DEEP LEARNING METHOD(2021) Pan, BoyangPurpose:
Deep learning has been widely applied in traditional medical imaging tasks like segmentation and registration. Some fundamental CNN based deep learning methods have shown great potential in low dose CT (LDCT) enhancement. This study applied U-Net++ model to enhance CT images with low dose and compared the performance of U-net++ and U-net quantitatively and qualitatively.
Method:
30 patient CT images were chosen as the ground truth in the training process. Under-sampled projections were simulated from the ground truth volumes with an uniform distribution. LDCT was then reconstructed from the under-sampled projections using the ASD-POCS TV algorithm with 40 iterations and was treated as the input of the models. The U-net++ model was improved based on U-net model by connecting the decoders, reserving better dense feature along skip connections. Deep supervision (DS) were used to make a combined loss between each upper node and the ground truth to enhance the image feature preserving capacity. U-net was used as standard model for comparison. L1 loss and structure similarity (SSIM) loss were used in different attempts. The generated images were compared quantitatively using SSIM and peak signal-to-noise ratio (PSNR).
Results:
Both models succeeded to improve the quality of the low dose CT images. The U-net++ model trained with MSE loss had best average PSNR of 17.8 on the test dataset and average SSIM of 0.779 in terms of the whole images compared with the original under-sampled LDCT with SSIM of 0.532 and PSNR of 16.7. U-net model trained using L1 loss had the best average SSIM of 0.756 and the average PSNR of 17.5. Conclusion: Deep learning method showed its potential dealing with the high dose caused by modern CT technique. Different CNN model could influence the quality of the generated image on different evaluation criterions.
Item Unknown Modelling the contributing factors of hypoxic fractions in human solid tumors(2021) Fan, YijiaHypoxia is one of the primary causes of radioresistance in human solid tumors. There are a series of contributing factors such as blood-flow rate, oxygen consumption rate, arterial pO2, and microvascular arrangement. In this study, blood-flow rate, oxygen consumption rate (OCR), and arterial pO2 are simulated to analyze their effects on hypoxia. A Green’s function method is applied in a densely vascularized tumor region to predict oxygen delivery. The results indicate that changes in OCR lead to greater changes in hypoxia. A 30% reduction in oxygen consumption rate leads to a 21% reduction in the hypoxic fraction. However, a 30% increase in blood-flow rate and arterial pO2 results in a 6% and 13% reduction in hypoxic fraction respectively. A 20% reduction in oxygen consumption rate plus a 20% increase in arterial pO2 causes a 40% decrease in hypoxic fraction. With increasing blood-flow rate and arterial pO2, hypoxic fraction reaches a plateau. As a result, hypoxia is more sensitive to OCR. In the second phase of this work, modeling results were compared to experimental data indicating the effect of papaverine in modulating tumor hypoxia. Papaverine is an FDA (Food and Drug Administration)-approved drug that can effectively decrease oxygen consumption rate and thus potentially decrease hypoxic fraction (~14%) at 2 mg/kg. At 4mg/kg, the addition of papaverine did not lead to a decrease in hypoxic fraction. This result fits the hypothesis that the oxygen consumption rate may be balanced by the effects of vasodilation which may induce more tumor shunting and poor perfusion. However, the results shown here are limited to few animals. A larger number of animals in each group, combined with other types of evidence such as perfusion staining is conducive to get a more accurate result and a better understanding of the underlying mechanism. Therefore, these experiments can be regarded as preliminary results and suggest opportunities for future experimental work.
Item Open Access Novel Designs of Radiomics-Integrated Deep Learning Models(2022) Hu, ZongshengPurpose: To investigate the feasibility of integrate radiomics and deep learning in computer-aided medical imaging analysis Methods: Two different approaches were investigated to integrate radiomics and deep leaning on two independent tasks respectively. In the first approach, a 2D sliding kernel was implemented to map the impulse response of radiomic features throughout the entire chest X-ray image; thus, each feature is rendered as a 2D map in the same dimension as the X-ray image. Based on each of the three investigated deep neural network architectures, including VGG-16, VGG-19, and DenseNet-121, a pilot model was trained using X-ray images only. Subsequently, 2 radiomic feature maps (RFMs) were selected based on cross-correlation analysis in reference to the pilot model saliency map results. The radiomics-boosted model was then trained based on the same deep neural network architecture using X-ray images plus the selected RFMs as input. The proposed radiomics-boosted design was developed using 812 chest X-ray images with 262/288/262 COVID-19/Non-COVID-19 pneumonia/healthy cases, and 649/163 cases were assigned as training-validation/independent test sets. For each model, 50 runs were trained with random assignments of training/validation cases following the 7:1 ratio in the training-validation set. Sensitivity, specificity, accuracy, and ROC curves together with Area-Under-the-Curve (AUC) from all three deep neural network architectures were evaluated. In the second approach, a cohort of 235 GBM patients with complete surgical resection was divided into short-term/long-term survival groups with 1-yr survival time threshold. Each patient received a pre-surgery multi-parametric MRI exam with 4 scans: T1, contrast-enhanced T1 (T1ce), T2, and FLAIR. Three tumor subregions were segmented by neuroradiologists, and the whole dataset was divided into training, validation, and test groups following a 7:1:2 ratio. The developed model comprises three data source branches: in the 1st radiomics branch, 456 radiomics features (RF) were calculated from the three tumor subregions of each patient’s MR images; in the 2nd deep learning branch, an encoding neural network architecture was trained for survival group prediction using each single MR modality, and high-dimensional parameters from the last two network layers were extracted as deep features (DF). The extracted radiomics features and deep features were processed by a feature selection procedure to reduce the dimension size of each feature space. In the 3rd branch, patient-specific clinical features (PSCF), including patient age and three tumor subregions volumes, were collected from the dataset. Finally, data sources from all three branches were fused as an integrated input for a supporting vector machine (SVM) execution for survival group prediction. Different strategies of model design were investigated in comparison studies, including 1) 2D/3D-based image analysis, 2) different radiomics feature space dimension reduction methods, and 3) different data source combinations in SVM input design. Results: In the first approach, all three investigated deep neural network architectures demonstrated improved sensitivity, specificity, accuracy, and ROC AUC results in COVID-19 and healthy individual classifications. VGG-16 showed the largest improvement in COVID-19 classification ROC (AUC from 0.963 to 0.993), and DenseNet-121 showed the largest improvement in healthy individual classification ROC (AUC from 0.962 to 0.989). The reduced variations suggested improved robustness of the model to data partition. For the challenging Non-COVID-19 pneumonia classification task, radiomics-boosted implementation of VGG-16 (AUC from 0.918 to 0.969) and VGG-19 (AUC from 0.964 to 0.970) improved ROC results, while DenseNet-121 showed a slight yet insignificant ROC performance reduction (AUC from 0.963 to 0.949). The achieved highest accuracy of COVID-19/Non-COVID-19 pneumonia/healthy individual classifications were 0.973 (VGG-19)/0.936 (VGG-19)/ 0.933 (VGG-16), respectively. In the second approach, the model achieved 0.638 prediction accuracy in the test set when using patient-specific clinical features only, which was higher than the results using radiomics features/deep features as sole input of SVM in both 2D and 3D based analysis. The inclusion of radiomics features or deep features with patient-specific clinical features improved accuracy results in 3D analysis. The most accurate models in 2D/3D analysis reached the highest accuracy of 0.745 with different combinations of dissimilarity-selected radiomics features, deep features, and patient-specific clinical features, and the corresponding ROC area-under-curve (AUC) results were 0.69 (2D) and 0.71 (3D), respectively.
Conclusions: The integration of radiomic analysis in deep learning model design improved the performance and robustness computer-aided diagnosis and outcome predication, which holds great potential for clinical applications and provides a radiomics perspective for deep learning interpretation.
Item Open Access Numerical Simulations for the Design of Novel Integrated Parallel Reception, Excitation, and Shimming (iPRES) Coil Arrays(2016) Wang, HongyuanMagnetic field inhomogeneity results in image artifacts including signal loss, image blurring and distortions, leading to decreased diagnostic accuracy. Conventional multi-coil (MC) shimming method employs both RF coils and shimming coils, whose mutual interference induces a tradeoff between RF signal-to-noise (SNR) ratio and shimming performance. To address this issue, RF coils were integrated with direct-current (DC) shim coils to shim field inhomogeneity while concurrently emitting and receiving RF signal without being blocked by the shim coils. The currents applied to the new coils, termed iPRES (integrated parallel reception, excitation and shimming), were optimized in the numerical simulation to improve the shimming performance. The objectives of this work is to offer a guideline for designing the optimal iPRES coil arrays to shim the abdomen.
In this thesis work, the main field () inhomogeneity was evaluated by root mean square error (RMSE). To investigate the shimming abilities of iPRES coil arrays, a set of the human abdomen MRI data was collected for the numerical simulations. Thereafter, different simplified iPRES(N) coil arrays were numerically modeled, including a 1-channel iPRES coil and 8-channel iPRES coil arrays. For 8-channel iPRES coil arrays, each RF coil was split into smaller DC loops in the x, y and z direction to provide extra shimming freedom. Additionally, the number of DC loops in a RF coil was increased from 1 to 5 to find the optimal divisions in z direction. Furthermore, switches were numerically implemented into iPRES coils to reduce the number of power supplies while still providing similar shimming performance with equivalent iPRES coil arrays.
The optimizations demonstrate that the shimming ability of an iPRES coil array increases with number of DC loops per RF coil. Furthermore, the z direction divisions tend to be more effective in reducing field inhomogeneity than the x and y divisions. Moreover, the shimming performance of an iPRES coil array gradually reach to a saturation level when the number of DC loops per RF coil is large enough. Finally, when switches were numerically implemented in the iPRES(4) coil array, the number of power supplies can be reduced from 32 to 8 while keeping the shimming performance similar to iPRES(3) and better than iPRES(1). This thesis work offers a guidance for the designs of iPRES coil arrays.