Among the most exciting areas of digital x-ray imaging physics over the last decade has been the development of cone-beam CT (CBCT) systems in a broad scope of applications ranging from diagnostic imaging to image-guided interventions. Enabled by the emergence of high-performance flat-panel x-ray detectors and fully 3D reconstruction techniques, the proliferation of CBCT raises new challenges in understanding fully 3D imaging performance (e.g., the fully 3D noise-power spectrum, noise-equivalent quanta, and task-based detectability) and new opportunities in a broad variety of innovative imaging platforms tailored to specific new applications and clinical tasks.

Dual-energy CT is rapidly proliferating in diagnostic and interventional imaging, offering improved material differentiation, soft-tissue visualization, an quantitative analysis with applications ranging from renal stone characterization to musculoskeletal imaging. Leveraging advanced 3D reconstruction methods and a rigorous understanding of signal and noise propagation in the 3D imaging chain, we have established a framework for DE-CBCT imaging performance analysis that highlights the factors governing tomographic performance in material decomposition / discimination and yields a basis for system optimization - e.g., selection of kVp pair and allocation of dose between high- and low-energy projections.

A general framework for modeling the 3D imaging chain has been realized through cascaded systems analysis of signal and noise propagation from the x-ray source through the detector system and reconstruction algorithm. Such models provide a powerful framework for system design and optimization and reveal complex tradeoffs among quantum noise, anatomical background, source-detector orbit, detector design, reconstruction algorithm, and radiation dose. An understanding of the factors governing 3D imaging performance provides a rigorous basis for system development and translation.


Two categories of detectors are integrated on I-STAR imaging benches. In addition to (energy-integrating) flat-panel detectors, recent work involves the incorporation of (energy-discriminating) photon counting detectors offering exciting new possibilities in high-performance CT, including high spatial resolution (50 micron detector pixel pitch), digital energy discrimination (spectral imaging), and low-dose imaging performance (zero noise readout). Photon counting detectors include Si strip modules scalable to dimensions appropriate to volumetric imaging of the extremities. Cascaded systems models are under develop to describe the signal and noise propagation, including spectral distortion effects, photon counting CT.


While prevalent metrics such as DQE and NEQ provide a rigorous means of system characterization, they fall short of answering a basic question: Can I see it? By combining such metrics with Fourier descriptors of the imaging task, the gap between system performance and real observer performance can be bridged in way that allows analysis and optimization of new imaging imaging systems from first principles of imaging physics. Recent work successfully demonstrates the ability of task-based cascaded systems analysis to include the effects of anatomical clutter, x-ray scatter, and the 3D reconstruction chain in predicting real observer response over a broad range of imaging conditions.


Task-based image quality models for CBCT, tomosynthesis, and dual-energy imaging offer a powerful basis for understanding the fundamental factors governing imaging performance. Among the key variabiles is radiation dose, and cascaded systems analysis provides a valuable framework for probing the low-dose limits of detectability. For example, the tradeoffs among number of views, total angular arc, anatomical clutter, and electronic noise are rendered clear within a task-based analysis of detectability.


Monte Carlo methods provide a powerful means of modeling the imaging chain in a manner that includes all aspects of imaging physics, ranging from x-ray scatter to non-idealities in detector response. The I-STAR Lab is working with collaborators at the US Food and Drug Administration (FDA) Center for Devices and Radiological Health (CDRH) in the development of fast, GPU-accelerated Monte Carlo methods for modeling cone-beam CT. Such research enables methods for accurate calculation of x-ray scatter and dose under conditions of broadly varying geometry, and beam characteristics in different CBCT applications.

The implementation of fast Monte Carlo methods on GPU enables highly accurate, patient-specific calculation of x-ray scatter and radiation dose. The resulting x-ray scatter corrections include complexities of tissue heterogeneity and various scatter components (e.g., Compton and Rayleight) that defy simple parametric corrections. The ability to compute the patient-specific (and tissue-specific) radiation dose imparted with each CT scan provides a far more complete dosimetric description beyond conventional descriptors of scanner output (CTDI) and accounts for various system geometries and source-detector orbits.
Within the I-STAR Laboratory is a facility for medical image reading studies - the Volume Image Visualization and Analysis (VIVA) Lab. The facility includes a collection of calibrated, diagnostic-quality displays and computer workstations integrated with the research network and clinical PACS. A dark-controlled reading environment is ideal for conducting reader studies under controlled conditions. The OPTEx software package (the "Observer Perfromance Test Engine") was developed in-house for streamlining image pre-processing, reading order randomization, observer training, response recording, and analyzing results in MAFC, preference, and ROC tests.


Known-component reconstruction (KCR) involves a novel approach to model-based 3D image reconstruction that simultaneously solves both a statistical / iterative reconstruction of the patient anatomy AND a registration of a known component (e.g., interventional device) somewhere within the image. The combined solution demonstrates enormous potential for improving image quality and eliminating artifacts that plague conventional reconstruction approaches that do not account for such strong prior information. KCR extensions include deformable components (dKCR) and statistically-known components (sKCR). Applications include imaging in the presence of surgical devices, implants, prosthesis, and more.


A general penalized likelihood (PL) framework has been developed for model-based 3D image reconstruction in cone-beam CT, providing a valuable basis for novel reconstruction techniques such as KCR, PIR-PLE, and MCR and providing dramatic improvement in image quality compared to conventional FBP under conditions of low dose and/or sparsely sampled data. Increasingly sophisticated forward models within the PL framework leverage a knowledge of the imaging chain gained from cascaded systems analysis of detector efficiency, blur, electronic noise, and other non-idealities conventionally ignored in model-based reconstruction.


A general framework combining model-based (PL) reconstruction and prior-image-based reconstruction methods is under development that exploits the benefits of BOTH explicit noise models (improving image quality at reduced radiation dose) AND prior information (e.g., patient-specific priors from previous scans). The combination is referred to as Prior-Image-based Penalized Likelihood Estimation (PI-PLE) with extensions including necessary Registration (incl. deformable registration) of the prior image (PIR-PLE) to avoid artifacts and false information. Compared to conventional model-based and prior-image-based approach, PIR-PLE offers benefits of low-dose imaging and full integration withf unregistered priors.


The general model-based framework (including PL, PIR-PLE, KCR, etc.) with GPU-accelerated implementations provide an important basis for investigating the low-dose limits of imaging performance in a broad spectrum of applications. In image-guided interventions, for example, where repeat CBCT are acquired during the procedure, the PL framework demonstrates dramatic image quality improvements at a fraction of the scan dose compared to FBP.


The ability to model the noise and spatial resolution in 3D imaging has been a major factor enabling the development of high-performance CBCT imaging systems over the last decade. As 3D reconstruction turns to nonlinear model-based methods, the ability to model image quality characteristics is equally important. Methods are under development for computing the spatially variant covariance matrix, noise, NPS, and spatial resolution characteristics in penalized likelihood reconstruction.


Prior-image-based reconstruction offers tremendous promise but raises questions as to whether features arise from the current data or from the prior images. We have developed a novel framework to analyze the propagation of information by decomposing the process into distinct components supported by the current data and by the prior image. This decomposition quantifies sources of information as a spatial map and traces specific features to their source. Such “information source maps” can be used as a check that a given feature arises from the current data and to quantitatively guide the selection of parameter values affecting the strength of prior information.


A limiting factor in some applications of CBCT is patient motion during fairly long scans (e.g., >20-60 s). Motion-compensation reconstruction (MCR) leverages a combination of model-based reconstruction (e.g., PL) for high image quality from sparse projection data and deformable image registration (e.g., variants of Demons) to build 4D motion vector fields. Whereas, conventional methods can suffer from low image quality in sparse data and errors in the motion estimate, a PL+Demons framework is shown to provide MCR that resolves image quality in the presence of motion, allows imaging at any desired motion phase, and yields high-quality 4D imaging.

The I-STAR Lab has developed a growing library of source code for the construction, application, and analysis of advanced reconstruction algorithms, image registration, and other sophisticated processing of 2D and 3D data. The tools are generally accessible from Matlab for straightforward inclusion in algorithm development efforts, and computationally intensive aspects of the toolbox are written in C/C++ and utilize CUDA libraries for computation on graphics cards. Tools and building blocks include: FDK, ML, PL (paraboloidal surrogates), PICCS, SART, and matched projectors / backprojectors (Siddon, distance-driven, and separable footprints).


Deformable image registration is essential to accurately aligning information in the reference frame of the most up-to-date point in an interventional procedure. Research in the I-STAR Lab has developed a framework for fast deformable registration that adapts methods such as the Demons algorithm to applications in image-guided radiation therapy and a spectrum of image-guided surgeries. Extensions include robust implementation within a morphological pyramid, an intensity-invariant form allowing CT-to-CBCT registration, and a novel super-dimensional form allowing registration in the presence of missing tissue and/or devices introduced between image pairs.


Deformable registration of the inflated and deflated lung is an important aspect of guiding minimally invasive trans-thoracic interventions (e.g., lung tumor surgery). We have developed a combined model-based (mesh and control points) registration and image-based (Demons) registration that is robust to large deformations and sufficiently accurate to guide the surgeon in relation to the target wedge, the tumor, and adjacent critical anatomy.


A novel approach to multi-dimensional deformable image registration (referred to as Extra-Dimensional Demons, XDD) is under development that exapnds the dimensionality of images in a manner that accoutns for tissue changes between the moving and fixed images (e.g., excised tissue or an interventional device in the latter). The algorithm explicitly models tissue discrepancies in in the process of tissue deformation. Excisions are accounted for by increasing the dimensionality of the solution so that deformations are represented by in-volume vectors while excisions are handled by out-of-volume vectors


The intensity profiles of CT and CBCT images can differ to CBCT intensity inaccuracy arising from large scatter fraction, different reconstruction algorithms, and other factors. For integration of preoperative CT with intraoperative CBCT, a variant of the Demons algorithm has been developed to include concurrent tissue-specific intensity matching that simultaneously performs registration and a iterative intensity match in order to improve registration accuracy and robustness against CBCT intensity inaccuracy.


The combination of imaging and surgical robotics presents a powerful synergy in precise, minimally invasive surgery. In trans-oral base-of-tongue surgery, the ability to deform preoperative data to the intraoperative scene allows the surgeon to fully exploit the precision possible in robot-assisted resection. A combined model-based and Demons-based algorithm for registering the preop (closed mouth) and intraop (open mouth, tongue disstended) scenes is under development to enable such approaches to trans-oral surgery.


Fast implementations of 3D-2D registration (e.g., CBCT-to-fluoroscopy) have been realized leveraging the power of multi-GPU architectures in combination with a general framework for various similarity metrics (e.g., NCC, MI, gradient information, etc.) and optimization schemes (e.g., downhil simplex, CMA-ES, etc.). Incorporation of 3D-2D registration within the TREK guidance system allows fast superposition of 3D data defined in CBCT in real-time fluoroscopy.


The development of high-performance intraoperative 3D imaging promises to overcome the limitations of conventional image-guided surgery with up-to-date images during surgery that accurately present changes associated with patient setup, tissue deformation, and target excision. A prototype mobile C-arm for high-quality cone-beam CT has been developed in collaboration with industry partners to answer such clinical challenges. The prototype provides volumetric images with sub-mm spatial resolution and soft-tissue visibility at low radiation dose and is the centerpiece of an integrated surgical navigation system undergoing translation to key surgical applications.


An imaging system dedicated to extremities CBCT has been developed for application in musculoskeletal radiology, orthopaedics, and rheumatology. Functionality includes weight-bearing (standing) or non-weight-bearing (sitting) configurations and combines digital radiography and real-time fluoroscopy with 3D imaging in a compact design. Research includes image quality optimization, minimizing radiation dose, advanced 3D reconstruction algorithms, and translation to clinical trials. Advanced applications include dual-energy cone-beam CT, imaging in the presence of metal implants (e.g., total knee replacement), 4D functional imaging, and sophisticated analysis of joint space morphology as an arthritis biomarker.


An experimental imaging bench in the I-STAR Lab provides a precise, flexible platform for investigation of imaging performance in flat-panel cone-beam CT and other advanced x-ray imaging modalities, such as tomosynthesis and dual-energy imaging. The bench is based upon an optical table with the x-ray source, detector, and imaging subject mounted on computer-controlled translation and rotation stages. The motion system is capable of emulating a broad range of x-ray system geometries and source-detector orbits with precise, reproducible geometry and a high degree of control over the acquisition process. It can be used not only for the investigation of imaging performance but also as a resource for preclinical studies that may benefit from high quality x-ray imaging.


A pre-clinical test-bed for development and optimization of mobile DE radiography for bedside applications. The testbed features a computer-controlled mobile x-ray unit and a high-performance, wireless flat-panel x-ray detector (Carestream DRX-1). The system provides an experimental setup for optimization of DE imaging protocols and minimizing radiation dose, with primary applications in the ICU aimed at enhanced visualization of lines and catheters.


The TREK software architecture was developed for integration of new surgical guidance technologies and acceleration to clinical use. The platform is based on proven open-source libraries, including CISST libraries for surgical tracking and 3D Slicer libraries for visualization and analysis. Modules are developed according to the surgical task, including: 3D deformable registration; 3D-2D registration to real-time fluoroscopy; surgical trackers used individually or in hybrid arrangements; augmented overlay of image and planning data in endoscopic video; real-time "virtual fluoroscopy" computed from GPU-accelerated DRRs; and multi-modality image display.


Augmentation of endoscopic video (including sinus endoscopy, thoracoscopy, laparoscopy, bronchoscopy, etc.) with preoperative or intraoperative image data can improve navigation, spatial orientation, surgical targeting, and avoidance of adjacent critical anatomy. In skull base surgery, the ability to overlay the carotid arteries, optic nerves, and tumor in real-time video exposes key structures beyond the visible surface and could offer a benefit to high-precision surgery. Similarly in thoracic surgery, the ability to highlight the target wedge and position of critical vessels and airways could greatly benefit targeting of subpalpable nodules in lung-conserving surgery.


The ability ot overlay preoperative or intraoperative image data within real-time video endoscopy requires accurate calibration of extrinsic and intrinsic camera parameters. A fast calibration method has been implemented in TREK sutiable for use by a trained OR technologist in clinical studies. The automated calibration includes a novel color-pattern checkerboard grid and correlation of eigenvalue-based features with iterative homography to solve for camera parameters. A novel hand-eye calibration employs compact dual quaternions for fast, efficient solution of translation and rotation using singular value decomposition, improving speed and accuracy in system setup.


A variety of new and commercially available surgical trackers are being applied to new applications, including infrared-based (Polaris), video-based (MicronTracker), and electromagnetic (EM) (Aurora) trackers. Research includes individual and concurrent use of trackers in configurations that are optimally suited to a given application. Hybrid tracker configurations leverage the advantages of each and overcome conventional limitations (e.g., line-of-sight and EM field distortion) and improve tracking precision over individual trackers. Such configurations require careful attention to streamlined implementation (to the point of invisibility) to augment surgical performance without complicating workflow.


A novel surgical tracking configuration is under development that positions the tracker directly on a rotational C-arm, improving geometric accuracy and line-of-sight in comparison to a conventional arrangement. Accurate registration is maintained within the rotating reference frame using a novel reference marker that can be sensed from any C-arm angle. Affixing the tracker also opens new functionality, such as virtual fluoroscopy (a simulated projection image from the current C-arm pose) and video augmentation (overlay of image data within the video scene) from a natural perspective over the surgical field.


Using a new electromagnetic (EM) tracker system developed by NDI (Mississauga ON) in which the field generator occupies a window-frame enclosure, we have developed a system in which the tracker is integrated within the operating table. The Tracker-in-Table improves field of view, resides outside the sterile field, maintains tracking over a large range of measurement, and is compatible with both fluoroscopy and CBCT. Methods to maintain tracker registration in the presence of a rotational C-arm have been developed, and the system can be integrated in a variety of hybrid in-room or on-C tracker configurations.


Surgical navigation requires registration of patient data (‘Image’) with the patient (‘World’). Conventional manual methods co-localize fiducial markers by mouse click and a handheld trackable pointer, which is time consuming and error-prone. A new automatic method has been developed using the high-performance CBCT C-Arm, featuring a robust Hough-Transform algorithm for detecting markers in CBCT projections. Performance is being evaluated: 1) in various anatomical sites (head, thorax, abdomen); 2) at various distances between markers and CBCT isocenter; 3) in comparison to conventional manual registration .


Next-generation image-guided spine surgery aims to incorporate high-performance intraoperative C-arm CBCT for improved image quality, soft-tissue visibility, and streamlined integration with guidance systems. Research includes translation of high-performance CBCT into the surgical suite in combination with a variety of application-specific navigation tools, advanced 3D reconstruction methods, fast deformable registration of preop and intraoperative images, novel tracking methods, and minimization of radiation dose to both the patient and surgical staff.


The proximity of surgical targets to critical structures in the skull base presents a challenge even to experienced surgeons. Limited visualization of the surgical field, anatomical deformation, and a high cost of complication highlight the need for high-quality intraoperative imaging. High-quality C-arm CBCT offers 3D images with sub-mm spatial resolution, soft-tissue visibility, and dose sufficiently low to permit repeat intraoperative imaging. Integration with preoperative data, tracking, and endoscopy offers a system for high-precision skull base surgery that could extend image guidance to a broad spectrum of head and neck pathologies.


Lung cancer remains the leading cancer killer, causing more deaths each year than the next two most common cancers (prostate and breast) combined. In its earliest stages, lung cancer is curable by surgery, and low-dose CT appears to offer a sensitive means of detecting lung tumors before they have spread. However, small tumors can be difficult to localize in surgery. C-arm cone-beam CT is being brought to bear in order to guide thoracic surgeons precisely to lung tumors in video-assisted thoracic surgery (VATS), potentially allowing surgeons to localize small tumors in the operating room and excise them precisely while sparing healthy lung tissue.


The combination of intraoperative CBCT and surgical robotics opens new possibilities for surgical precision in minimally invasive interventions. Research collaboration between I-STAR, LCSR, and clinical programs at Johns Hopkins Hospital is seeking to synergize imaging and surgical robotics. For example, trans-oral robotic surgery (TORS) of base-of-tongue tumors augmented by intraoperative CBCT, deformable image registration, and augmentation of stero endoscopy is under development offers to improve both surgical targeting and patient safety.


The ability to acquire real-time 2D and 3D in guidance of needle biopsy and minimally invasive ablative therapies offers improved targeting precision and enhanced patient safety. Systems under development include multi-modality fluoroscopy and CBCT on the same C-arm platform combined with fast 3D-2D registration for enhanced visualization and workflow.


The early development of in-room volumetric image guidance based on cone-beam CT was among the early work of I-STAR investigators at William Beaumont Hospital, the University of Toronto, and at Johns Hopkins University. Implementation of volumetric imaging on a linear accelerator provided a powerful new tool for accommodating motion of the tumor and surrounding organs in high-precision radiation therapy. Ongoing work addresses issues of soft-tissue image quality, low-dose imaging techniques, and 3D deformable image registration.


A prototype CBCT scanner for musculoskeletal (MSK) radiology opens new possibilities for imaging of the extremities. The system demonstrates superior spatial resolution and lower radiation dose in comparison to conventional MDCT with improved cost, footprint, and workflow. Capabilities under developmetn include: imaging the weight-bearing knee, ankle, and foot for diagnosis of impingement; combined radiography / fluoroscopy / CBCT on a single platform; functional 2D and 4D imaging; and dual-energy CBCT for material discrimination beyond conventional CT contrast limits.


The capabilities for isotropic, sub-mm spatial resolution combined with soft-tissue visibility (e.g., cartilage, tendon, ligament, etc.) suggest an important role for the dedicated extremity CBCT scanner. Fine visualization of bone details open new avenues for fracture monitoring and understanding mechanisms in fracture healing. Quantitative imaging of bone mineral density offers a new modality for quantitative CT and osteoporosis imaging. Visualization of muscles, tendons, and ligaments could augment surgical planning and therapy assessment.


In addition to evaluation of cartilage and erosions in osteoarthritis (OA) imaging, the prototype CBCT scanner offers a new tool in the arsenal against rheumatoid arthritis (RA). The scanner offers a modality improved diagnosis, staging, and treatment evaluation (e.g., in disease-modifiying drug therapies) by fine visualization of soft-tissue involvement (synovium, cartilage, and bone marrow edema), high-resolution imaging of erosions and subchondral architecture, and quantitative image-based findings of early, middle, and late RA findings.


Imaging protocols and calibration techniques are under development on the dedicated extremity CBCT scanner to provide volumetric quantitative CT (QCT) measurements of bone mineral density (BMD). Such capability could provide a valuable addition to the diagnostic arsenal in imaging of osteoporosis, monitoring of therapy response, and analysis of fracture risk.


As CT technology has evolved rapidly over the last decade to include multiple-source and fully 3D volumetric beams, methods for accurate dose measurement and calculation require commensurate advances. Development of the various x-ray imaging modalities within the I-STAR Lab is guided by rigorous scrutiny and minimization of radiation dose. Engineers, physicists, and clinicians work in collaboration to ensure the safe application of CT in research and patient care. I-STAR researchers are also engaged in several efforts throughout the scientific community to standardize dosimetry methods and promote techniques that minimize CT dose.


Dual-energy (DE) imaging allows discrimination of specific materials within the body - for example, calcium, iodine, metal, and various soft-tissues - and offers increased conspicuity and quantitation of subltle disease. A theoretical framework for DE imaging system optimization has been applied to the creation prototype DE radiography and tomography systems. Through identification of optimal kVp pair, beam filtration, dose allocation, advanced decomposition techniques, and translation to clinical studies, we demonstrated improved diagnostic performance in comparison to single-energy imaging techniques without increase in dose.


Visualizing the placement of interventional devices can be challenged by poor contrast in conventional mobile radiography. Dual-energy (DE) images acquired using a bedside mobile radiography system demonstrates a major improvement in visualization of bone, soft-tissue, and plastic / metal device components. The system uses a portable, wireless, flat-panel detector (Carestream DRX-1) feasible for deployment in bedside exams in the ICU. The increased conspicuity of tubes, lines, catheters, needles, and stents, offers to improve the speed and quality of bedside imaging, reduce the number of image retakes, and improve patient safety.


Among the most common forms of wrong-site surgery is wrong-level spinal intervention. The challenge arises from the lack of uniquely identifiable landmarks in the fluoroscopic scene, particularly in the mid-thoracic spine. A fast 3D-2D regstration technique ("LevelCheck") has been developed to automatically register data defined in preoperative CT (specifically, definition of vertebral levels) and overlay labels directly in real-time fluoroscopy as a means of assistance and decision support in vertebral targeting.


Intraoperative CBCT imaging protocols are identified that minimize the radiation dose in a manner sufficient for the imaging task - e.g., visualization of bone detail versus soft-tissue structures. Advanced model-based statistical reconstruction methods are translated to early studies that further reduce the low-dose limits of detectability. For example, PL reconstruction tuned to bone or soft-tissue visualization reduces CBCT dose such that the total dose delivered over the course of a procedure is less than the dose of a single diagnostic CT scan.


Surgical navigation systems entail a level of geometric error associated with the accuracy of the tracker, image-to-world calibration, and arrangment of registration fiducials. Analytical methods that identify optimal calibration methods and fiducial configurations are under development. The option to communicate geometric errors within the surgical navigation system is under investigation as a means to facilitate more knowledgeable approach in high-precision surgery. For example, we are investigating the effect on surgical precision, workflow, and decision-making when the target registration error (TRE) is exposed graphically within the navigation system.


High-quality intraoperative CBCT offers not only a method for surgical guidance in high-precision surgery but also a method for detecting complications in the OR immediately following surgery (which might otherwise be detected hours later in post-operative CT). The detection of intracranial hemorrhage, for example, could offer a major quality check in neurosurgery. Similarly, the ability to detect breach of the central nervous system in sinus surgery could provide a valuable check against CSF leak. Such capabilities provide important safety checks in the OR when the surgeon can best undertake revisions against potential complications.


Technologies for advanced visualization provide a potentially important means of improving patient safety and surgical workflow. In spine surgery, for example, localizing a target vertebra by conventional methods can entail a significant amount of time and radiation dose (e.g., fluoroscopic level counting). The abiliity to overlay image data directly on a video scene (e.g., Tracker-on-C or endoscopic video) augments the surgeon's view of target anatomy and adjacent critical anatomy, potentially streamlining target localization to single view and facilitating lower-dose, higher precision minimially invasive surgery.


High-quality intraoperative imaging provides a means of not only guiding high-precision surgery, but also a means for quantitative record, verify, documentation, and analysis of the surgical product. Tools such as KCR provide quantitative analysis of surgical device placement in comparison to intended trajectories. Intraoperative CBCT is under investigation as a means of surgical product evaluation by quantitative metrics of target ablation / resection and normal tissue avoidance. Standardized image-based and quantitative reporting of the surgical product facilitates longer-term goals in large population-based studies aimed to understand variabilities in treatment response.


The I-STAR Laboratory
Department of Biomedical Engineering | Johns Hopkins University
Traylor Building, Room 726 | 720 Rutland Avenue | Baltimore, MD 21205