- Interventional Radiology
Honors & Awards
Moskowitz Research Award, Stanford Radiology Residency (2020)
Resident Research Grant, Society of Interventional Radiology (2020)
Resident/Fellow Travel Award, Society of Interventional Oncology (2020)
Mary Ellis Bell Prize Research Paper Prize Competition, Perelman School of Medicine at University of Pennsylvania (2017)
RSNA-APSA Introduction to Academic Radiology Traveling Award, RSNA-APSA (2016)
Traveling Award, Antibody Engineering & Therapeutics Conference, Antibody Engineering & Therapeutics Conference (2015)
Paul and Daisy Soros Fellowship for New Americans, Paul and Daisy Soros Foundation (2013)
PennAPPS Programming Hackathon, University of Pennsylvania (2013)
PennVention 2013 Competition Third Place, University of Pennsylvania (2013)
Weiss Tech House Innovation Fund Recipient, University of Pennsylvania School of Engineering (2013)
Wharton Business Plan Competition Semifinalist, Wharton School at University of Pennsylvania (2013)
World Molecular Imaging Congress Conference Travel Award, World Molecular Imaging Congress (WMIC)/ World Molecular Imaging Society (WMIS) (2013)
Wharton Medtronic Penn Case Competition, 1st place, Wharton School at University of Pennsylvania (2012)
Howard Hughes Medical Institute -NIBIB Biomedical Imaging Interface Program Fellowship, Howard Hughes Medical Institute -NIBIB (2009-2017)
Howard Hughes Undergraduate Research Scholar, UCLA-Howard Hughes Medical Institute (2007-2009)
M.D., Perelman School of Medicine at University of Pennsylvania, Medicine (2017)
Ph.D., University of Pennsylvania, Bioengineering (2015)
B.S., University of California Los Angeles, Biochemistry and Biomedical Research (2009)
Practical computer vision application to detect hip fractures on pelvic X-rays: a bi-institutional study.
Trauma surgery & acute care open
2021; 6 (1): e000705
Pelvic X-ray (PXR) is a ubiquitous modality to diagnose hip fractures. However, not all healthcare settings employ round-the-clock radiologists and PXR sensitivity for diagnosing hip fracture may vary depending on digital display. We aimed to validate a computer vision algorithm to detect hip fractures across two institutions' heterogeneous patient populations. We hypothesized a convolutional neural network algorithm can accurately diagnose hip fractures on PXR and a web application can facilitate its bedside adoption.The development cohort comprised 4235 PXRs from Chang Gung Memorial Hospital (CGMH). The validation cohort comprised 500 randomly sampled PXRs from CGMH and Stanford's level I trauma centers. Xception was our convolutional neural network structure. We randomly applied image augmentation methods during training to account for image variations and used gradient-weighted class activation mapping to overlay heatmaps highlighting suspected fracture locations.Our hip fracture detection algorithm's area under the receiver operating characteristic curves were 0.98 and 0.97 for CGMH and Stanford's validation cohorts, respectively. Besides negative predictive value (0.88 Stanford cohort), all performance metrics-sensitivity, specificity, predictive values, accuracy, and F1 score-were above 0.90 for both validation cohorts. Our web application allows users to upload PXR in multiple formats from desktops or mobile phones and displays probability of the image containing a hip fracture with heatmap localization of the suspected fracture location.We refined and validated a high-performing computer vision algorithm to detect hip fractures on PXR. A web application facilitates algorithm use at the bedside, but the benefit of using our algorithm to supplement decision-making is likely institution dependent. Further study is required to confirm clinical validity and assess clinical utility of our algorithm.III, Diagnostic tests or criteria.
View details for DOI 10.1136/tsaco-2021-000705
View details for PubMedID 33912689
View details for PubMedCentralID PMC8031685
Practical computer vision application to detect hip fractures on pelvic X-rays: a bi-institutional study
Trauma Surgery Acute Care Open
View details for DOI 10.1136/tsaco-2021-000705
Diagnostic performance of lower extremity Doppler ultrasound in detecting iliocaval obstruction.
Journal of vascular surgery. Venous and lymphatic disorders
To retrospectively evaluate the performance of two commonly used Doppler ultrasound parameters, namely, venous flow phasicity and response to Valsalva maneuver, in detecting iliocaval obstruction.All imaging studies of patients seen by interventional radiology for lower extremity venous disease at a single institution from 1996 to 2018 were retrospectively identified. Lower extremity ultrasounds with a concurrent magnetic resonance, computed tomography, or conventional venogram performed within the next 7 days, which served as gold standard, were further identified (n = 192 examinations, including 313 limbs). Iliocaval obstruction were assessed by two ultrasound criteria: (1) nonphasic flow and/or (2) nonresponsive flow to Valsalva in the common femoral vein. The sensitivity, specificity, negative predictive value (NPV) and positive predictive value (PPV) for diagnosing iliocaval obstruction were calculated for each ultrasound criterion, and also for when the two criteria were assessed jointly.Of the 313 limbs assessed for venous flow phasicity, 133 (42.5%) had an iliocaval obstruction confirmed on subsequent venography. Nonphasic flow demonstrated a sensitivity of 69.2%, specificity of 82.8%, NPV of 78.4%, and PPV of 74.8% for diagnosing iliocaval obstruction. Of the 212 limbs assessed for Valsalva response, 88 (41.5%) had a confirmed iliocaval obstruction. Nonresponsive flow to Valsalva demonstrated a sensitivity of 13.6%, specificity of 97.6%, NPV of 61.6%, and PPV of 80.0% for diagnosing iliocaval obstruction. Joint assessment using phasicity and Valsalva criteria demonstrated a sensitivity of 68.2%, specificity of 87.2%, NPV of 79.6%, and PPV of 78.9%.In this tertiary care setting, Doppler ultrasound examination was not a reliable diagnostic tool for detecting iliocaval obstruction.
View details for DOI 10.1016/j.jvsv.2019.12.074
View details for PubMedID 32107162
Quantitative Control of Gene-Engineered T-Cell Activity through the Covalent Attachment of Targeting Ligands to a Universal Immune Receptor.
Journal of the American Chemical Society
2020; 142 (14): 6554–68
Universal immune receptors represent a rapidly emerging form of adoptive T-cell therapy with the potential to overcome safety and antigen escape challenges faced by conventional chimeric antigen receptor (CAR) T-cell therapy. By decoupling antigen recognition and T-cell signaling domains via bifunctional antigen-specific targeting ligands, universal immune receptors can regulate T-cell effector function and target multiple antigens with a single receptor. Here, we describe the development of the SpyCatcher immune receptor, the first universal immune receptor that allows for the post-translational covalent attachment of targeting ligands at the T-cell surface through the application of SpyCatcher-SpyTag chemistry. The SpyCatcher immune receptor redirected primary human T cells against a variety of tumor antigens via the addition of SpyTag-labeled targeting ligands, both in vitro and in vivo. SpyCatcher T-cell activity relied upon the presence of both target antigen and SpyTag-labeled targeting ligand, allowing for dose-dependent control of function. The mutational disruption of covalent bond formation between the receptor and the targeting ligand still permitted redirected T-cell function but significantly compromised antitumor function. Thus, the SpyCatcher immune receptor allows for rapid antigen-specific receptor assembly, multiantigen targeting, and controllable T-cell activity.
View details for DOI 10.1021/jacs.9b11622
View details for PubMedID 32191035
View details for PubMedCentralID PMC7306176
In Vivo Effects of Retrobulbar Bimatoprost Injection on Orbital Fat
LIPPINCOTT WILLIAMS & WILKINS. 2018: 201–4
Recent publications have reported the adverse effects of prostaglandin analogues on the periocular tissues. These medications may cause periorbital lipodystrophy, enophthalmos, and deepening of the superior sulcus deformity. While these effects may have adverse consequences for some patients, the atrophy of the periorbital fat may have a useful role in diseases that lead to orbital and periorbital fat hypertrophy such as thyroid eye disease. In this pilot study, the authors investigated the effects of retrobulbar bimatoprost injection on the intraocular pressure and orbital fat in a rat animal model.Three rats were sedated and intraocular pressure was measured. A 0.1 ml aliquot of bimatoprost was injected into the right orbit of all rats. In the left orbit, 0.1 ml of phosphate-buffered saline was injected as a control. Three weeks later, all rats were sedated and intraocular pressure was measured before euthanizing. Routine histologic staining was performed and thin sections through the intraconal orbital fat were obtained. Density of intraconal adipocytes was measured and adipocyte heterogeneity was determined using a computer image analysis algorithm.The specimens injected with bimatoprost demonstrated atrophy of orbital fat with significantly increased adipocyte density (p = 0.009) and heterogeneity (p = 0.008) when compared with control. Intraocular pressure was not significantly decreased at 3 weeks after injection of retrobulbar bimatoprost.In this pilot study, orbital injection of bimatoprost demonstrated atrophy of intraconal adipocytes when compared with control orbits injected with saline. The orbits injected with bimatoprost were noted to have smaller, more heterogeneous adipocytes that were densely packed in the intraconal space. The study limitations include the small sample size, which limited the ability for us to make conclusions about the effect on intraocular pressure. Nevertheless, the findings presented suggest that retrobulbar bimatoprost may present a nonsurgical alternative to induce atrophy of the orbital fat without inducing inflammation or hypotony.
View details for PubMedID 28369019
View details for PubMedCentralID PMC5623620
- Site-specific antibody-liposome conjugation through copper-free click chemistry: a molecular biology approach for targeted photodynamic therapy SPIE-INT SOC OPTICAL ENGINEERING. 2016
Biodistribution, Clearance, and Toxicology of Polymeric Micelles Loaded with 0.9 or 5 nm Gold Nanoparticles
JOURNAL OF BIOMEDICAL NANOTECHNOLOGY
2015; 11 (10): 1836–46
Long-circulating gold nanoparticles (AuNPs) have garnered a great deal of interest as both imaging and therapeutic agents. However, their protracted elimination and long-term persistence within many organ systems remains a concern for clinical translation. To improve the excretion of long-circulating nanoparticles, we prepared -80 nm biodegradable polymeric micelles with 0.9 nm or 5 nm AuNPs tightly packed within the hydrophobic core. These gold-loaded polymeric micelles (GPMs) were expected to allow for improved excretion of gold, compared with single large AuNPs, owing to the smaller size and larger surface-to-volume ratio of the individual AuNPs within the micelle. Following intravenous administration of GPMs, organs were harvested and examined for gold content using inductively coupled plasma optical emission spectrometry (ICP-OES) for up to 3 months post-injection. While both GPM formulations showed significant clearance of gold over time, micelles containing 0.9 nm AuNPs showed a 72% and 67% reduction in gold content in the liver and spleen, respectively, between 1 day and 3 months post-injection, compared with a 38% and 35% reduction in mice receiving 5 nm GPMs. Furthermore, feces and urine analysis revealed approximately 7.5 and 100 times more gold, respectively, in mice that received 0.9 nm GPMs one day after injection. These findings suggest that the excretion profile of inorganic nanomaterials may be improved if clusters of small inorganic materials are used in favor of single solid particles.
View details for DOI 10.1166/jbn.2015.2142
View details for Web of Science ID 000359391800013
View details for PubMedID 26502646
View details for PubMedCentralID PMC4942304
LASIC: Light Activated Site-Specific Conjugation of Native IgGs
2015; 26 (8): 1456–60
Numerous biological applications, from diagnostic assays to immunotherapies, rely on the use of antibody-conjugates. The efficacy of these conjugates can be significantly influenced by the site at which Immunoglobulin G (IgG) is modified. Current methods that provide control over the conjugation site, however, suffer from a number of shortfalls and often require large investments of time and cost. We have developed a novel adapter protein that, when activated by long wavelength UV light, can covalently and site-specifically label the Fc region of nearly any native, full-length IgG, including all human IgG subclasses. Labeling occurs with unprecedented efficiency and speed (>90% after 30 min), with no effect on IgG affinity. The adapter domain can be bacterially expressed and customized to contain a variety of moieties (e.g., biotin, azide, fluorophores), making reliable and efficient conjugation of antibodies widely accessible to researchers at large.
View details for PubMedID 26057140
Optimization of Photoactive Protein Z for Fast and Efficient Site-Specific Conjugation of Native IgG
2014; 25 (9): 1709–19
Antibody conjugates have been used in a variety of applications from immunoassays to drug conjugates. However, it is becoming increasingly clear that in order to maximize an antibody's antigen binding ability and to produce homogeneous antibody-conjugates, the conjugated molecule should be attached onto IgG site-specifically. We previously developed a facile method for the site-specific modification of full length, native IgGs by engineering a recombinant Protein Z that forms a covalent link to the Fc domain of IgG upon exposure to long wavelength UV light. To further improve the efficiency of Protein Z production and IgG conjugation, we constructed a panel of 13 different Protein Z variants with the UV-active amino acid benzoylphenylalanine (BPA) in different locations. By using this panel of Protein Z to cross-link a range of IgGs from different hosts, including human, mouse, and rat, we discovered two previously unknown Protein Z variants, L17BPA and K35BPA, that are capable of cross-linking many commonly used IgG isotypes with efficiencies ranging from 60% to 95% after only 1 h of UV exposure. When compared to existing site-specific methods, which often require cloning or enzymatic reactions, the Protein Z-based method described here, utilizing the L17BPA, K35BPA, and the previously described Q32BPA variants, represents a vastly more accessible and efficient approach that is compatible with nearly all native IgGs, thus making site-specific conjugation more accessible to the general research community.
View details for PubMedID 25121619
Facile Method for the Site-Specific, Covalent Attachment of Full-Length IgG onto Nanoparticles
2014; 10 (16): 3354–63
Antibodies, most commonly IgGs, have been widely used as targeting ligands in research and therapeutic applications due to their wide array of targets, high specificity and proven efficacy. Many of these applications require antibodies to be conjugated onto surfaces (e.g. nanoparticles and microplates); however, most conventional bioconjugation techniques exhibit low crosslinking efficiencies, reduced functionality due to non-site-specific labeling and random surface orientation, and/or require protein engineering (e.g. cysteine handles), which can be technically challenging. To overcome these limitations, we have recombinantly expressed Protein Z, which binds the Fc region of IgG, with an UV active non-natural amino acid benzoylphenyalanine (BPA) within its binding domain. Upon exposure to long wavelength UV light, the BPA is activated and forms a covalent link between the Protein Z and the bound Fc region of IgG. This technology was combined with expressed protein ligation (EPL), which allowed for the introduction of a fluorophore and click chemistry-compatible azide group onto the C-terminus of Protein Z during the recombinant protein purification step. This enabled the crosslinked-Protein Z-IgG complexes to be efficiently and site-specifically attached to aza-dibenzocyclooctyne-modified nanoparticles, via copper-free click chemistry.
View details for PubMedID 24729432
View details for PubMedCentralID PMC4142076
VOLUME AND COMPOSITION OF REFLUX AFTER INTRAVITREAL INJECTION
RETINA-THE JOURNAL OF RETINAL AND VITREOUS DISEASES
2014; 34 (7): 1473–76
To quantify the amount of drug loss from cadaveric human eyes, which are injected via the pars plana with a known volume of dye at variable intraocular pressures.Eight cadaver eyes were divided into 2 intraocular pressure groups: normal (15 mmHg; 4 eyes) or high (30 mmHg; 4 eyes). Each eye was injected with 50 μL of hematoxylin dye, and the subsequent reflux was immediately collected on a Schirmer's test strip. The test strip was scanned and digitally analyzed to determine the area of saturation and total color intensity present. Using a previously established equation, total volume of reflux and the amount of dye within that reflux were calculated.The average total volume of refluxed fluid was 1.68 μL (median, 0.62 μL), with a range of 0 μL to 8.05 μL. The average volume of refluxed dye was 0.37 μL (median, 0.08 μL), with a range of 0 μL to 2.15 μL. On average, only 0.74% of the original 50 μL of injected dye was lost (median, 0.15%), with a range from 0% to 4.30%.Although the presence of subconjunctival bleb formation after intravitreal injection may be a concern to the clinician, data from the present study shows that only a very small amount of the injected therapeutic agent is lost in the reflux.
View details for PubMedID 24451925
View details for PubMedCentralID PMC4065616
A Novel Method for the Measurement of Reflux from Intravitreal Injections: Data from 20 Porcine Eyes
CURRENT EYE RESEARCH
2014; 39 (7): 752–57
Reflux following intravitreal injection is a common phenomenon, but it is unknown how much, if any, medication is lost as a result. Reflux is known to be a combination of vitreous and the injected agent, but the relative composition is unknown. This article describes a novel method for the measurement of the volume and composition of reflux and presents data from porcine eyes.Twenty porcine eyes were injected with 0.05 ml of dye at intraocular pressures (IOPs) of 15, 20, 25 and 30 mmHg (five eyes per subgroup). Reflux was captured on filter paper and the area of saturation and color intensity of the dye were digitally analyzed. Total refluxed volume and proportion of dye versus vitreous fluid were calculated from linear regression lines created from known standards.Average (median) total volume of reflux from all eyes was 1.19 μl (0.93 μl), volume of injected dye refluxed was 0.47 μl (0.11 μl) and composition of reflux was 20.8% dye (15.5%). Less than 1% of the injected dye was lost to reflux. There were no differences between IOP groups in the total volume refluxed, the total amount of dye refluxed, the average composition of the reflux or the amount of injected dye refluxed (df = 3 for all comparisons; p = 0.58, p = 0.51, p = 0.55, p = 0.51, respectively).This novel method allows for measurement of quantity and composition of reflux following intravitreal injection in vitro. While reflux occurs frequently, it is predominantly composed of vitreous, not the injected agent. In fact, <1% of the original injection was lost to reflux.
View details for PubMedID 24256543
View details for PubMedCentralID PMC4028430
- Research Highlights: highlights from the latest articles in nanomedicine. Nanomedicine (London, England) 2014; 9 (4): 385–88
Multifunctional Nanoparticles: Cost Versus Benefit of Adding Targeting and Imaging Capabilities
2012; 338 (6109): 903–10
Nanoparticle-based drug delivery systems have been developed to improve the efficacy and reduce the systemic toxicity of a wide range of drugs. Although clinically approved nanoparticles have consistently shown value in reducing drug toxicity, their use has not always translated into improved clinical outcomes. This has led to the development of "multifunctional" nanoparticles, where additional capabilities like targeting and image contrast enhancement are added to the nanoparticles. However, additional functionality means additional synthetic steps and costs, more convoluted behavior and effects in vivo, and also greater regulatory hurdles. The trade-off between additional functionality and complexity is the subject of ongoing debate and the focus of this Review.
View details for PubMedID 23161990
View details for PubMedCentralID PMC3660151
Simultaneous Quantification of Tumor Uptake for Targeted and Nontargeted Liposomes and Their Encapsulated Contents by ICPMS
2012; 84 (17): 7578–82
Liposomes are intensively being developed for biomedical applications including drug and gene delivery. However, targeted liposomal delivery in cancer treatment is a very complicated multistep process. Unfavorable liposome biodistribution upon intravenous administration and membrane destabilization in blood circulation could result in only a very small fraction of cargo reaching the tumors. It would therefore be desirable to develop new quantitative strategies to track liposomal delivery systems to improve the therapeutic index and decrease systemic toxicity. Here, we developed a simple and nonradiative method to quantify the tumor uptake of targeted and nontargeted control liposomes as well as their encapsulated contents simultaneously. Specifically, four different chelated lanthanide metals were encapsulated or surface-conjugated onto tumor-targeted and nontargeted liposomes, respectively. The two liposome formulations were then injected into tumor-bearing mice simultaneously, and their tumor delivery was determined quantitatively via inductively coupled plasma mass spectroscopy (ICPMS), allowing for direct comparisons. Tumor uptake of the liposomes themselves and their encapsulated contents was consistent with targeted and nontargeted liposome formulations that were injected individually.
View details for PubMedID 22882145
Targeted and drug-loaded micelles
2012; 7 (7): 951–52
View details for PubMedID 23019673
Improving nanoparticle delivery with anti-angiogenesis therapy
2012; 7 (7): 949–50
View details for PubMedID 22846089
Synergizing radiation therapy with drug-loaded nanovectors
2012; 7 (7): 950–51
View details for PubMedID 23019671
Enhancing the cell uptake of nanoparticles within tumors
2012; 7 (7): 951
View details for PubMedID 23019672
Probability fold change: A robust computational approach for identifying differentially expressed gene lists
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
2009; 93 (2): 124–39
Identifying genes that are differentially expressed under different experimental conditions is a fundamental task in microarray studies. However, different ranking methods generate very different gene lists, and this could profoundly impact follow-up analyses and biological interpretation. Therefore, developing improved ranking methods are critical in microarray data analysis. We developed a new algorithm, the probabilistic fold change (PFC), which ranks genes based on a confidence interval estimate of fold change. We performed extensive testing using multiple benchmark data sources including the MicroArray Quality Control (MAQC) data sets. We corroborated our observations with MAQC data sets using qRT-PCR data sets and Latin square spike-in data sets. Along with PFC, we tested six other popular ranking algorithms including Mean Fold Change (FC), SAM, t-statistic (T), Bayesian-t (BAYT), Intensity-Conditional Fold Change (CFC), and Rank Product (RP). PFC achieved reproducibility and accuracy that are consistently among the best of the seven ranking algorithms while other ranking algorithms would show weakness in some cases. Contrary to common belief, our results demonstrated that statistical accuracy will not translate to biological reproducibility and therefore both quality aspects need to be evaluated.
View details for DOI 10.1016/j.cmpb.2008.07.013
View details for Web of Science ID 000263313600002
View details for PubMedID 18842321
Innate immunity and transcription of MGAT-III and Toll-like receptors in Alzheimer's disease patients are improved by bisdemethoxycurcumin
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA
2007; 104 (31): 12849–54
We have tested a hypothesis that the natural product curcuminoids, which has epidemiologic and experimental rationale for use in AD, may improve the innate immune system and increase amyloid-beta (Abeta) clearance from the brain of patients with sporadic Alzheimer's disease (AD). Macrophages of a majority of AD patients do not transport Abeta into endosomes and lysosomes, and AD monocytes do not efficiently clear Abeta from the sections of AD brain, although they phagocytize bacteria. In contrast, macrophages of normal subjects transport Abeta to endosomes and lysosomes, and monocytes of these subjects clear Abeta in AD brain sections. Upon Abeta stimulation, mononuclear cells of normal subjects up-regulate the transcription of beta-1,4-mannosyl-glycoprotein 4-beta-N-acetylglucosaminyltransferase (MGAT3) (P < 0.001) and other genes, including Toll like receptors (TLRs), whereas mononuclear cells of AD patients generally down-regulate these genes. Defective phagocytosis of Abeta may be related to down-regulation of MGAT3, as suggested by inhibition of phagocytosis by using MGAT3 siRNA and correlation analysis. Transcription of TLR3, bditTLR4, TLR5, bditTLR7, TLR8, TLR9, and TLR10 upon Abeta stimulation is severely depressed in mononuclear cells of AD patients in comparison to those of control subjects. In mononuclear cells of some AD patients, the curcuminoid compound bisdemethoxycurcumin may enhance defective phagocytosis of Abeta, the transcription of MGAT3 and TLRs, and the translation of TLR2-4. Thus, bisdemethoxycurcumin may correct immune defects of AD patients and provide a previously uncharacterized approach to AD immunotherapy.
View details for DOI 10.1073/pnas.0701267104
View details for Web of Science ID 000248603900047
View details for PubMedID 17652175
View details for PubMedCentralID PMC1937555