Browsing by Author "Tsekos, Nikolaos V."
Now showing 1 - 20 of 61
- Results Per Page
- Sort Options
Item 3D Reconstruction of Tubular Structures Using MRI Projection Images(2018-05) Unan, Mahmut 1986-; Tsekos, Nikolaos V.; Shah, Dipan J.; Leiss, Ernst L.; Shi, WeidongAfter imaging information became available in digital form, techniques for acquiring volumetric data evolved. 3D reconstruction is mostly performed using multislice stack images. The objective of this dissertation is to introduce a simple magnetic resonance technique for imaging tubular structures, such as blood vessels and catheters, and 3D reconstruction of these structures. This study includes three major chapters: one on simulation and two on experiments with MRI projection images. First, a MATLAB simulation was created to analyze the reconstruction process; it was tested with different shapes of the structures and different numbers of projections. Second, triplanar projection imaging was evaluated on a phantom filled with a T1-shortening, Gd-based contrast agent embedded into a lipid matrix. The object is reconstructed from three mutually orthogonal projections of the volume that contain the structure of interest. The projected structures of the object were segmented out on each projection, back-projected to generate the segmented tubular object, and mesh-rendered in 3D. The accuracy of this approach was investigated by comparing the mesh-rendered tubular structure generated from projections with the mesh rendered from a multislice set of images of the same volume. Third, Inverse Radon Transform was implemented for 3D reconstruction of complex helical tubular structure from multiple radially deployed (oblique) projections. To compute the correctness of the 3D reconstruction processes, we compared the resulting meshes with the multislice-rendered meshes. Hausdorff distance and Point Cloud Comparison methods were used to evaluate the reconstruction error. The average error was less than 1 pixel for the triplanar projection images, and it was less than 2 pixels for the oblique orientation projection images. With further optimization and reduction of acquisition time, this method can be used for 3D fast imaging of interventional tools or segments of blood vessels with applications in interventional MRI.Item A Computational Framework to Understand Vascular Adaptation(2015-05) Rahman, Mahbubur 1982-; Garbey, Marc; Berceli, Scott A.; Tsekos, Nikolaos V.; Gabriel, Edgar; Hilford, VictoriaResearchers have been continuously applying a wide variety of approaches to understand vascular adaptation over the past two decades. However, the specific cause/effect or links between the hemodynamic factors, inflammatory biochemical mediators, cellular effectors and vascular occlusive phenotype remain unexplained still today. To explain these biological phenomena, we have introduced a multi-scale computational framework to systematically test many hypotheses associated with the vascular adaptation and finally applied this framework to explain some widely observed clinical and experimental cases. Our framework incorporates the cellular activities inside the vein graft influenced by the shear stress and tension, which are two of the most important environmental factors in the vascular adaptation. This is a hybrid agent based model (ABM) coupled with the partial differential equations (PDEs) associated with the calculation of the shear stress. Based on the computational framework, we have designed and developed a modular, adaptive, efficient and scalable simulation program so that we can explain some specific pattern formations associated with the vascular adaptation by pattern recognition algorithms of the framework in real time. Finally, we have coupled a genetic algorithm with the framework to verify the fact that a combination of interesting patterns associated with the vascular adaptation can be regenerated in a multivariate data analysis environment. As a result, this research will reduce the gap in understanding different cases observed in the vascular adaptation.Item A Computational Image-Based Guidance System for Precision Laparoscopy(2016-12) Nguyen, Toan B. 1984-; Tsekos, Nikolaos V.; Pavlidis, Ioannis T.; Vilalta, Ricardo; Garbey, MarcThis dissertation presents our progress toward the goal of building a computational image-based guidance system for precision laparoscopy; in particular, laparoscopic liver resection. As we aim to keep our working goal as simple as possible, we have focused on the most important questions of laparoscopy - predicting the new location of tumors and resection plane after a liver maneuver during surgery. Our approach was to build a mechanical model of the organ based on pre-operative images and register it to intra-operative data. We proposed several practical and cost-effective methods to obtain the intra-operative data in the real procedure. We integrated all of them into a framework on which we could develop new techniques without redoing everything. To test the system, we did an experiment with a porcine liver in a controlled setup: a wooden lever was used to elevate a part of the liver to access the posterior of the liver. We were able to confirm that our model has decent accuracy for tumor location (approximately 2 mm error) and resection plane (1% difference in remaining liver volume after resection). However, the overall shape of the liver and the fiducial markers still left a lot to be desired. For further corrections to the model, we also developed an algorithm to reconstruct the 3D surface of the liver utilizing Smart Trocars, a new surgical instrument recognition system. The algorithm had been verified by an experiment on a plastic model using the laparoscopic camera as a mean to obtain surface images. This method had millimetric accuracy provided the angle between two endoscope views is not too small. In an effort to transit our research from porcine livers to human livers, in-vivo experiments had been conducted on cadavers. From those studies, we found a new method that used a high-frequency ventilator to eliminate respiratory motion. The framework showed the potential to work on real organs in clinical settings. Hence, the studies on cadavers needed to be continued to improve those techniques and complete the guidance system.Item A Feasible Method to Evaluate RF-Induced Heating Behavior of Passive Orthopedic Implants in Human Body(2019-05) Guo, Ran; Chen, Ji; Zouridakis, George; Tsekos, Nikolaos V.; Jackson, David R.; Kainz, WolfgangMagnetic resonance imaging (MRI) radio frequency (RF) -induced heating is one of the most important concerns of MRI safety for patients with implantable medical devices. Due to the difficulties of performing direct study of the RF-induced heating in human body for an implantable medical device, the current standard method is to investigate the RF-induced heating in a standard and fully controlled phantom, which is filled with a homogeneous media to mimic the human body tissues. However, the RF-induced heating in a homogeneous regular shaped media is different from that in a heterogeneous human body, especially patients with orthopedic healthcare products. Numerical studies were already conducted to illustrate the difference of RF-induced heating on medical plates between phantom and human body. It is necessary to study the intrinsic mechanism of RF-induced heating in heterogeneous human body with passive implantable medical device and evaluate it with a feasible accurate method. Numerical modeling and simulations are conducted to study the RF-induced heating for typical orthopedic implants, such bone plates, hip prosthesis, and tibia intramedullary nails, in 1.5T and 3T magnetic resonance (MR) environment. Comparison results of RF-induced heating between phantom and human body are conducted to show the disadvantages of phantom. In order to study the mechanism of RF-induced heating for passive orthopedic implants, Huygens source and heterogeneous phantom are applied which could illustrate the effect of incident field and medium, respectively. Additionally, a homogeneous human-shape phantom is applied to study the RF-induced heating for each orthopedic implant. The effect of medium electric properties on the incident electric field distribution inside phantom structure are investigated. And local low lossy medium is added to mimic the effect of human bone tissue on RF-induced heating. Based on these results and analysis, a new phantom structure is proposed to properly evaluate the RF-induced heating behavior of passive orthopedic implants. Compared to traditional ASTM phantom method, the new phantom structure could achieve the exact RF-induced heating properties for passive orthopedic implants.Item A Framework for Interactive Immersion into Imaging Data Using Augmented Reality(2022-04-25) Velazco, Jose D.; Tsekos, Nikolaos V.; Leiss, Ernst L.; Eick, Christoph F.; Navkar, Nikhil V.; Webb, Andrew G.Image acquisition scanners produce an ever-growing amount of 3D/4D multimodal data that requires extensive image analytics and visualization of collected and generated information. For the latter, augmented reality (AR) with head-mounted displays (HMDs) has been commended as a potential enhancement. This PhD describes a framework (FI3D) for interactive and immersive experiences using an AR interface powered with image processing and analytics. The FI3D was designed to communicate with peripherals, including imaging scanners and HMDs, and to provide computational power for data acquisition and processing. The core of the FI3D is deployed in a dedicated unit that executes the computationally demanding processes in real-time; the HMD is used as an IO interface of the system. The FI3D is customizable and allows users to integrate different workflows while incorporating third party libraries. Using the FI3D as a foundation, two applications were developed in the cardiac and urology medical domains to experiment with, test, and validate the system. First, cine MRI images were segmented using a machine learning model while simultaneously an HMD rendered the reconstructed surfaces. Secondly, a simulated environment for robotic assisted MRI-guided transrectal prostate biopsies was developed, and user studies were conducted to evaluate the feasibility of AR visualization and interaction using the HoloLens HMD. Performance results showed that the system can maintain an image stream of five images with a resolution of 512 x 512 per second and update visual properties of the holograms at 1 update per 16 milliseconds. Interactive studies showed that using a gaming joystick allowed the manipulation of a robotic structure more effectively than using holographic menus or a mouse and keyboard. The FI3D can serve as the foundation for medical applications that benefit from AR visualization, removing various technical challenges from the developmental pipeline. The versatility, immersive, and interactive experience offered by the AR interface may assist physicians with diagnosis and image-guided interventions, resulting in safer and faster procedures. This can further increase the accessibility of healthcare to the public, yielding an increase in patient throughput.Item A Multiscale Model for Breast-Conservative Therapy: Computational Framework and Clinical Validation(2015-08) Simonetti, Valentina 1991-; Garbey, Marc; Bass, Barbara L.; Tsekos, Nikolaos V.Breast cancer is the most common cancer among women worldwide and affects 12% of all the women in the USA. There exist different surgical approaches in order to defeat this kind of cancer: the traditional mastectomy (Breast Removal Surgery) and the more recent Breast-Conservative Therapy (BCT), whose goal is to preserve the breast contour and ameliorate the psycological impact of surgery on the patients. This work aims to exploit the BCT field developing a 3D patient-specific multiscale model that could predict the breast shape after lumpectomy, from surgery to complete healing. This model consists of two parts: a hyperelastic Neo-Hookean Finite-Element Model of the breast tissues and skin, and a Cellular Automata model that mimics the biology of healing after surgery. The resulting multiscale model shows results that agree with our theoretical assumptions and gives as outcome the breast contour after surgery depending on the anatomy of the patient and the input from the surgeon. This work is, in fact, the result of an interdisciplinary collaboration between surgeons, mathematicians and computer scientists. A clinical protocol that involves patients eligible for BCT was developed in order to validate this multiscale model with clinical data. The results obtained show the performance of the model and our findings based on the data of the first patient who took part of the study. The model validation gave us an error of maximum 2.5 cm for the surface comparison, which implies the need of further improvements. The Cellular Automata model showed fairly accurate results with the preliminary data, but we need more patients in order to obtain conclusions that are statistically consistent.Item A Multiuser, Multisite, and Platform-Independent On-the-Cloud Framework for Interactive Immersion in Holographic XR(2024-03-01) Neeli, Hosein; Tran, Khang Q.; Velazco-Garcia, Jose Daniel; Tsekos, Nikolaos V.Background: The ever-growing extended reality (XR) technologies offer unique tools for the interactive visualization of images with a direct impact on many fields, from bioinformatics to medicine, as well as education and training. However, the accelerated integration of artificial intelligence (AI) into XR applications poses substantial computational processing demands. Additionally, the intricate technical challenges associated with multilocation and multiuser interactions limit the usability and expansion of XR applications. Methods: A cloud deployable framework (Holo-Cloud) as a virtual server on a public cloud platform was designed and tested. The Holo-Cloud hosts FI3D, an augmented reality (AR) platform that renders and visualizes medical 3D imaging data, e.g., MRI images, on AR head-mounted displays and handheld devices. Holo-Cloud aims to overcome challenges by providing on-demand computational resources for location-independent, synergetic, and interactive human-to-image data immersion. Results: We demonstrated that Holo-Cloud is easy to implement, platform-independent, reliable, and secure. Owing to its scalability, Holo-Cloud can immediately adapt to computational needs, delivering adequate processing power for the hosted AR platforms. Conclusion: Holo-Cloud shows the potential to become a standard platform to facilitate the application of interactive XR in medical diagnosis, bioinformatics, and training by providing a robust platform for XR applications.Item A Novel Transmission Mechanism for MRI-Compatible Surgery Robot(2017-12) Liu, Xin 1976-; Tsekos, Nikolaos V.; Becker, Aaron T.; Chen, Guoning; Shi, WeidongMinimally invasive surgery (MIS) techniques provide reduced patient discomfort, faster healing time, decreased risk of complications, and better overall patient outcomes. Medical imaging guidance is particularly crucial for MIS in which the procedure is performed through small openings in the body which resulting in limited sensory information available to surgeon compared with the open approach. Magnetic Resonance Imaging (MRI) is an intrinsically three-dimensional (3D) modality which offers high contrast and spatial resolution and a plethora of soft-tissue contrast mechanisms for assessment of anatomical morphology and function. These benefit , in addition to the fact that it does not require ionizing radiation, makes it a desirable methodology for image-guided interventions (IGI). An impediment to those advancements, however, is the limited access to patients, especially to the high-fi eld cylindrical magnetic resonance (MR) scanners. To address the limited accessibility and facilitate real-time guidance of interventions, remotely actuated and controlled MRI-compatible manipulators have been introduced. The MRI compatible interventional systems require appropriate forms of actuation. The commonly used electromagnetic actuators by many robotic system, like surgical robots, are, in general, not compatible with the MRI environment owing to their magnetically susceptible materials and electromagnetic components which are MR unsafe. In this work, we propose a novel transmission mechanism, herein referred to as Solid-Media Transmission (SMT), to transmit force from MR unsafe components located outside of the MR environment to the end effectors which are MR safe or MR conditional. We focus on the design, fabrication and control of a SMT-based actuator and an integrated robotic system that are aimed to perform the task of a surgical tool or needle placement. Experimental studies have demonstrated the feasibility and characteristics of the SMT and SMT-enabled devices for MRI guided intervention.Item A PATTERN RECOGNITION APPROACH TO LEARNING TRACKS OF HEAVY-ION PARTICLES IN TIMEPIX DETECTORS(2013-12) Hoang, Son M. 1985-; Vilalta, Ricardo; Pinsky, Lawrence S.; Shah, Shishir Kirit; Johnsson, Lennart; Tsekos, Nikolaos V.The rapid development in semiconductor detector technology at CERN has provided the capability to develop an active personal dosimeter for use in space radiation environments. The work reported here is based on the Timepix chip, which when coupled with an Si sensor, can function as an active nuclear emulsion, allowing the visualization of the individual tracks created as the different incident particles traverse the detector. The Timepix chip provides the capability of measuring the energy deposited by each incident particle that traverses the sensor layer. Together with the capability for online readout, this detector opens the door to a completely new generation of active Space Radiation Dosimeters. Although recent advances in hardware technology promise a major step forward in the development of such active portable space radiation dosimeters, little effort has been devoted toward software tools for analysis and classification of sources of radiation. Coupling radiation dosimeter hardware with pattern recognition techniques and machine learning tools has the potential to greatly improve current applications on space dosimeter projects. Our focus is not only to measure dosimetric endpoints directly such as dose-equivalent, but also to determine the physical nature of the radiation field with sufficient precision to allow characterization of the radiation composition and energy spectrum.Item A Study of Visual Attention on Gestures and Motion during Infancy(2018-05) Nasir, Irteza 1990-; Shah, Shishir Kirit; Tsekos, Nikolaos V.; Yoshida, HanakoIn recent years, understanding development of children’s visual attention with the help of computer vision techniques have been promising. Many approaches have been tried to understand what are the factors that generate attention in infants. Analyzing videos taken from different perspectives have been increasingly useful in such studies as they provide new insights. Nevertheless, analyzing these videos frame by frame is time consuming and unmanageable. Moreover, it is difficult for humans to assess all of the parameters that impact child's visual attention. In this thesis, we have proposed a tool for extracting and analyzing the motions from videos of child-parent toy play. We have focused primarily on the third perspective videos. The approach first extracts dense trajectories from these videos, and then uses unsupervised clustering to group the trajectories into multiple groups. These groups are then analyzed to explore potential correlations between the motions of the parents and the attention of the child. The proposed tool will enable researchers to look into unknown patterns that might contribute into the development of children’s visual attention by analyzing child-parent toy play videos.Item A Transmission Line Model for Active Implantable Medical Devices under MRI(2018-05) Liu, Jingshen; Chen, Ji; Jackson, David R.; Chen, Jiefu; Tsekos, Nikolaos V.; Benhaddou, Driss; Kainz, WolfgangThe use of magnetic resonance imaging (MRI) is widely restricted for patients with active implantable medical devices (AIMDs) due to safety concerns. In the United States alone, there are millions of people having AIMDs and a significant number of them need follow-up MRI scanning during their lifetimes. Some of the major MRI safety concerns come from the interaction between the electromagnetic fields generated by the MRI radio-frequency (RF) coil and AIMDs. Such interaction can result in RF-induced heating in human tissues, which can cause tissue damage, and RF-induced voltages on devices, which can cause device damage or malfunction. When analyzing the MRI RF safety problems, full-wave electromagnetic simulation approaches are often computationally prohibitive due to the complexity of the structure of the AIMDs, and in vivo measurement approaches are often not feasible. Therefore, an electromagnetic model is necessary to simplify the problem. In this dissertation, a transmission line model is proposed to describe the electromagnetic properties of AIMDs. The model is derived using simplified leads and generalized to practical AIMD leads. Numerical and experimental validations were performed for the model. A method to use a simplified equivalent solid lead to replace the complex AIMD lead is also proposed. The method significantly reduces the computational cost of simulating AIMD leads, and it enables the full-wave simulation of an AIMD together with a human model and an MRI RF coil. Based on the transmission line model, various studies were performed on the assessment and improvement of the MRI RF safety for patients with AIMDs. The analytical relation between the MRI RF safety and impedances in the transmission line model is derived and validated. The relation shows that adding properly chosen lumped elements or filters at the two ends of an AIMD can significantly improve the MRI RF safety. The influences of various physical parameters on the transmission line parameters are analyzed. An approach to search for the best tissue simulating medium is proposed, and an approach to determine the MRI RF-induced voltage transfer function scaling factor is proposed.Item ALGORITHMS AND DATA STRUCTURES TO DETECT ONCOVIRUSES IN HUMAN CANCER USING NEXT GENERATION SEQUENCING DATA(2012-12) Zhu, Rui 1980-; Fofanov, Yuriy; Widger, William R.; Tsekos, Nikolaos V.Evidence suggests human cancer can be induced by viruses. One way to test this hypothesis is to look for viral sequences in the human cancer genome. Next Generation Sequencing (NGS) technology sequences the whole human genome in a short period of time. This opens a door for a systematic analysis of the human genome and a thorough search for oncogenic viral sequences in cancer. However, a huge amount of sequencing reads generated by NGS poses a great challenge on the computational part of data analysis in terms of computing speed and memory usage. Data structures such as hash and tree are widely implemented to improve the performance of computing algorithms. Here, I described both data structures that have been developed in our center and compared their performance. Hash out performed tree when mapping the reads to a small reference sequence database. Subsequently, real human cancer data were analyzed by using the hash-based mapper and different oncoviral sequences were found in different cancers.Item Algorithms for Particle Swarms Using Global Control: Aggregation, Mapping, Coverage, Foraging, and Shape Control(2017-05) Viswanathan Mahadev, Arun; Becker, Aaron T.; Tsekos, Nikolaos V.; Pan, MiaoTargeted drug delivery is a promising technique to reduce the side effects of drugs by delivering them in concentrated doses using large swarms (10^16) of controllable microbots only targeting bad or infected tissue. A promising way to control small steerable microbots is by using a global control field such as the magnetic gradient of an MRI machine. In this work we develop benchmark algorithms for performing aggregation of microbots using global control. Using our findings we develop algorithms for a novel approach of mapping tissue and vascular systems without the use of harmful contrast agents in an MRI. In our work we consider a swarm of particles in a 1D, 2D, and 3D grids that can be tracked and controlled by an external agent thus building a map. We present algorithms for controlling particles using global inputs to perform: (1) Mapping, i.e., building a representation of the free and obstacle regions of the workspace; (2) Foraging, i.e., ensuring that at least one particle reaches each target location;and (3) Coverage, i.e., ensuring that every free region on the map is visited by at least one particle. Finally we also demonstrate shape control of large swarms using global control by developing an algorithm for position control.Item An Efficient Hemodynamic Workflow in Computational Surgery(2013-08) Tran Son Tay, Guillaume Linh 1983-; Garbey, Marc; Berceli, Scott A.; Davies, Mark G.; Hilford, Victoria; Tsekos, Nikolaos V.For few decades, it has been shown that atherosclerosis is the cause of the majority of clinical cardiovascular diseases including peripheral arterial diseases. The diagnosis and treatment for vascular disease has evolved significantly over the past years considering the rapid advances in imaging technologies. In recent years, computational fluid dynamics has been increasingly used as a simulation tool for blood flows. Numerous researches connect wall shear stress quantities to endovascular diseases such as stenosis, aneurism, and atherosclerosis. A thorough knowledge of vascular anatomy and hemodynamic would be beneficial for understanding the development and progression of the disease, the therapeutic decision process and follow up. The objective of this dissertation is to propose a computational fluid dynamic framework that includes: Understanding how streamline efficiently hemodynamic simulation for main arteries to produce database for clinical study/Providing some confidence estimate on numerical results/Extending the state of the art of clinical study by including motion and particles analysis.Item Association Rule Mining for Risk Assessment in Epidemiology(2016-08) Toti, Giulia 1986-; Vilalta, Ricardo; Lindner, Peggy; Price, Daniel M.; Tsekos, Nikolaos V.In epidemiology, a risk assessment measures the association between exposures and a health outcome. Risk characterization has traditionally been performed using statistical methods such as logistic regression, but such methods are not effective when working with highly correlated variables and when trying to assess synergic actions between exposures. These limitations become evident in studies related to asthma, a common chronic that affects 25 million people in the US. The prevalence of asthma is growing and research is struggling to find the reason. Many factors have been associated with causing and triggering asthma, but their interactions, as well as which one is the most responsible for the spreading of asthma, are still unclear. Outdoor air pollution is on the list of possible causes and triggers. Characterizing the connection between asthma and air pollution is not an easy task, because of high collinearity between pollutant agents, possible synergic actions, and difficulty in controlling the exposure. The research community is currently encouraging the use of multi-pollutant models to yield better results. In this dissertation we propose: (i) a modified Apriori association rule mining method for identification of connections between exposures and risk variations, and (ii) a novel genetic algorithm (GA) designed to mine risk-based quantitative association rules. Both methods were tested on a group of synthetic datasets, and on real data collection about pediatric asthma cases and pollution levels in Houston. The results on the synthetic datasets show the advantages of applying our methods to augment traditional logistic regression, and help determining the best metrics to include in the GA fitness function (odds ratio, length, repetition and redundancy). Tests on clinical data suggest the existence of a correlation between asthma and outdoor air pollutants, both alone and as a mixture. The genetic algorithm improves the results of the Apriori-based method by recognizing what appear to be the most dangerous levels of exposure. Future work will help to improve aspects of the GA such as population initialization or rule selection. To date, the proposed methods represent a significant step in the direction of risk assessment based on association rule mining in epidemiological studies.Item Blockchain and Digital Signatures for Digital Self-Sovereignty(2018-12) Patel, Brijesh B. 1993-; Shi, Weidong; Tsekos, Nikolaos V.; Bronk, ChrisPrinciples of self-sovereignty have been integrated into the solution to achieve a mechanism where the user is in control of one's digital identity attributes. Through the use of attribute-based credentials, the solution presented here allows the user to control access to their digital identity attributes, so they only have to release the required attributes to the business entities. Selective disclosure proofs, enabled by cryptographically signed containers, allow for minimization of identity attributes transferred to execute a transaction. The user can consent to access of one's identity attributes by granting access licenses to business entities through a blockchain application running on their mobile device. Also, the user can modify the access license to restrict the access based on time or revoke access to any identity attribute. Privacy of identity attributes and access licenses stored on mobile devices is ensured by integration of transparent data encryption. Dependency on any middleman entity required by several other identity management solutions is eliminated through the use of digital signatures. The communication between actors involved in each transaction is encrypted through a PKI infrastructure ensuring the security of claims packages transferred. The solution enables portability through use of digital signature to verify the validation of identity attributes done by the identity guarantor. The user is able to determine the lifespan of any identity attribute through the mobile application and remove it from any future digital transaction. The solution presented here allows for the application of theoretical principles of self-sovereign identity into the everyday life of the user.Item Choosing the Right Kernel A Meta-Learning Approach to Kernel Selection in Support Vector Machines(2015-05) Valerio Molina, Roberto 1983-; Vilalta, Ricardo; Eick, Christoph F.; Tsekos, Nikolaos V.; Shi, Weidong; Kaiser, KlausIn recent years Support Vector Machines (SVM) have gained increasing popularity over other classification algorithms due to their ability to produce a flexible boundary over non-linearly separable datasets. Such an ability is feasible thanks to the kernel trick. The kernel trick allows SVMs to perform an implicit transformation of the non-linearly separable original input space into a higher dimensional feature space where a linear separation of the dataset can be found. By creating an implicit transformation of the original feature space we gain efficiency in terms of time complexity. However, we lose information since we do not know what the feature space looks like, but we obtain relative positions in the feature space thanks to the kernel function used to perform this transformation. Since different kernel functions yield different transformations of the feature space, there is a need for a mechanism that selects the best kernel function for a specific problem. Previous work has focused on generating metrics from the kernel matrix (a pairwise matrix that stores the relative positions of all the pairs of points). Three metrics have been used to extract information from the kernel matrix: Fisher's discriminant, Bregman's divergence and Homoscedasticity analysis, which even when combined together do not provide enough prediction power to perform kernel selection. By introducing new meta-features, Distance Ratio (capturing inter-class and intra-class distances in the feature space) and Class Similarity (computing inter-class and intra-class similarity in the feature space), we yield substantial improvements to the kernel selection process.Item Classification of 4D fMRI Images Using ML, Focusing on Computational and Memory Utilization Efficiency(2021-12) Beheshti, Nazanin; Johnsson, Lennart; Vilalta, Ricardo; Tsekos, Nikolaos V.; Nguyen, Hien VanAlzheimer’s disease is progressively degenerative with a character of memory loss, mood and behavior changes, and deepening confusion about time and place. It is estimated that worldwide about 50 million people are affected by AD. The lifetime per patient care cost of AD is estimated to be about $250k and the total cost of care of AD patients could exceed $1 trillion by 2050. In this research, we use novel data reduction techniques in determining functional brain connectivity from Resting-State fMRI data and show that small Machine Leaning models can with good accuracy classify subjects with respect to Alzheimer’s disease (AD) or Mild Cognitive Impairment (MCI) or being Cognitive Normal (CN). In fMRI, brain activity is captured from Blood Oxygen Level-Dependent (BOLD) magnetization detected by the MRI scanner. The functional connectivity is inferred from correlations of the observed BOLD signals from typically cubic voxels with sides in the 3 – 4 mm range. The BOLD signals are typically sampled every 2 – 3 seconds for a duration of five to six minutes generating a data set of 5 – 10 million voxel BOLD signal values per subject. To reduce the computational effort classification is typically carried out based on signal aggregates for anatomical regions defined in brain atlases. In this research, we use the 90 region Automated Anatomical Labeling atlas, AAL-90, in establishing Regions of Interest, ROIs that are subsets of voxels in the AAL-90 atlas. The functional connectivity is measured by the correlation of BOLD signal aggregates for the ROIs. In the data reduction step, we represent the 4D data set for a region with a vector that on average reduces the data set for a region from about 100,000 voxel signal values to 100 to 200 values in our spatial representation and in the order of 15,000 – 30,000 in our spatial-temporal representation. We show that a small Convolutional Neural Network (CNN) with a model size of about 168 kiB and a Transformer model of only 37 kiB yields classification accuracies of 80 – 90% for AD, MCI, and CN subject classification. We further show that our region data aggregation technique is more robust to BOLD signal artifacts than the commonly used aggregation technique. The training time for the CNN and Transformer on a data set of 551 subjects required 184 and 23.73 seconds respectively. The experiments are conducted on Opuntia Cluster using Pytorch.1.5.0, Python 3.7.7, and CUDA 10.1 on a 2.8GHz Intel Xeon E5-2670v2 processor with 2 CPU sockets and 20 cores, and NVIDIA K40 GPU.Item Computational Approaches to Detect Pathogens in the Presence of Complex Backgrounds(2012-12) Rojas, Mark 1973-; Fofanov, Yuriy; Widger, William R.; Chapman, Barbara M.; Tsekos, Nikolaos V.; Shah, Shishir KiritFast and accurate identification of pathogenic microorganisms in complex clinical and environmental samples is essential for the prevention and treatment of infectious diseases. The most sensitive and accurate detection approaches are based on the examination of the nucleic acid composition of the sample in order to identify the presence of pathogens DNA and/or RNA. A large spectrum of nucleic acid-based tests (such as PCR, RT-PCR, and oligonucleotide microarrays) is designed to examine a sample for the presence of pre-defined genomic signatures: short pathogen-specific DNA and/or RNA fragments. Identification of such signatures however, represents significant computational challenges. To be pathogen specific, each signature (or combination of signatures) must be present (conserved) across all strains of the pathogen, and absent in all other organisms including its close neighbors, and must have assay specific biochemical and thermodynamic properties, such as binding energy, melting temperature, and nucleotide composition. All available signature design algorithms rely on heuristics and are known to miss cases when potential signatures are (explicitly or with small number of mismatches) also present in host (human) and/or non-pathogen microorganisms causing false positive outcomes. Even greater challenge for the design of biochemical platform specific genomic signatures (probes and primers) is that each type of instrument uses different biochemical protocols to detect signatures which also have to be included in the consideration during the signatures design process. To address these challenges we have developed novel algorithms and data structures able to bring all possible subsequences located in given pathogen genome into signatures design process. Moreover, the developed algorithms make it possible to consider mismatches (insertions, deletions, and substitutions for all positions and combinations) into the design process. We also have developed the concept of ultra-specific genomic islands: genomic regions in which every subsequence is several mismatches away from the closest subsequence which may appear in a host genome and/or non-pathogenic near-neighbors of targeted pathogen. This concept allows to improve the quality and flexibility (genomic islands can be used to identify thermodynamically acceptable signatures) of the design of biochemical platform specific detection tests. Developed approach was successfully used to design a variety of tests for Category A, B, and C, pathogens including the 2009 H1N1 Influenza outbreak originated in Mexico.Item Computational Methods for MRI-Guided and Powered Ferric Applicators: Modeling and Image Processing(2021-12) Chu, Wenhui; Tsekos, Nikolaos V.; Shi, Weidong; Eick, Christoph F.; Becker, Aaron T.Cardiac diseases are major causes of global mortality which are a consistent threat to the lives of people. With the development of left ventricle segmentation, the real-time MRI-based control of a ferromagnetic application for endovascular navigation with data sensing and feedback in cardiac was applied in recent years. In this work, we first propose three novel deep learning architectures called BNU-net, LNU-net, and IBU-net for left ventricle segmentation from short-axis cine MRI images. BNU-net is the batch normalized (BN) U-net, LNU-net is the layer normalized (LN) U-net, and IBU-net is the instance-batch normalized (IB) U-net. The architectures of BNU-net, LNU-net, and IBU-net have an encoding path for feature extraction and a decoding path that enables precise localization. BNU-net, LNU-net, and IBU-net have left ventricle segmentation methods: BNU-net employs batch normalization to the results of each convolutional layer, LNU-net applies layer normalization in each convolutional block, while IBU-net incorporates instance and batch normalization together in the first convolutional block. Our method incorporates affine transformations and elastic deformations for image data processing. Our dataset that contains 805 MRI images regarding the left ventricle from 45 patients is used for evaluation. The experimental results reveal that our approach accomplishes comparable or better performance than other state-of-the-art approaches in terms of the dice coefficient and the average perpendicular distance. We then simulate a computational platform for preoperative planning and modeling of MRI-powered applicators inside blood vessels. This platform was implemented as a two-way data and command pipeline that links the MRI scanner, the computational core, and the operator. The platform first processes multi-slice MR data to extract the vascular bed and then fits a virtual corridor inside the vessel. This corridor serves as a virtual fixture (VF), a forbidden region for the applicators to avoid vessel perforation or collision. The geometric features of the vessel centerline, the VF, and MRI safety compliance (dB/dt, max available gradient) are then used to generate magnetic field gradient waveforms. Different blood flow profiles can be user-selected, and those parameters are used for modeling the applicator's maneuvering.