Graduate Student Research

Biomedical Engineering Posters | Awards

The Marquette University and Medical College of Wisconsin Joint Department of Biomedical Engineering supports a vast array of research initiatives. While each project is regularly discussed in terms of overarching objectives and long-term goals, sometimes the strokes that bring the project to fruition remain in the shadows. Every year, the Medical College of Wisconsin hosts a Graduate and PhD Research Poster session in which faculty, students, postdocs and staff are invited to gather, view, present, and discuss the finer points of some of their ongoing projects.  In recognition of the vital contribution each presented line of inquiry plays in the success of our initiatives, research posters presented by graduate students of the Joint Department of Biomedical Engineering have been collected here. 

Biomedical Engineering Posters


Headshot of Ruth WoehlkeArbitrary-Andle Split-Detection Using a Rotational Dove Prism in Adaptive Optics Scanning Light Ophthalmoscopy

This poster was presented by Ruth E. Woehlke under the direction of Dr. Robert F. Cooper

  • Poster (Click to open as PDF)
  • Abstract
  • Authors

Woelhke Research Poster

An adaptive optics scanning light ophthalmoscope was enhanced with a Dove prism to obtain arbitrary angle split-detection images. This application allows for a low-cost enhancement of extant systems and facilitates the resolution of retinal features.

Ruth E. Woehlke, Mina Gaffney, Ching Tzu Yu*, Hannah M. Follett*, Chloe Guillaume*, Joseph Carroll, Robert F. Cooper


Headshot of Dayeong AnMyocardial displacement field generation from CMR cine images by image-to-image translation deep-learning networks

This poster was presented by Dayeong An under the direction of Dr. El-Sayed Ibrahim

  • Poster (Click to open as PDF)
  • Abstract
  • Authors

Dayeong An Research Poster 2022

CMR cine and tagging sequences are typically used for quantifying global and regional cardiac functions, respectively. Nevertheless, the need for acquiring additional tagged images and using special software packages for tagged images analysis contributes to longer and more expensive CMR exam. The CMR feature tracking technique has been proposed to generate regional cardiac function measures from cine images, although the lack of intramyocardial markers poses a limitation of the technique compared to conventional CMR tagging. In this study, we developed a deep neural network algorithm for extracting regional cardiac function parameters from the cine images after the network is trained on corresponding gold-standard tagged images. Methods: The developed algorithm is based on image-to-image translation using generative adversarial network (GAN)1 to generate the myocardial displacement fields from cine images. During the training phase, the inputs of the network are: 1) cine difference images, generated by subtracting consecutive cine images to illustrate the endo- and epicardium motion; and 2) corresponding tagged images acquired of the same slice and cardiac phases. The target image of the network is the corresponding gold-standard myocardial displacement field generated by analyzing the tagged image using the SinMod method2,3. We used a dataset of 1134 input images acquired from rats scanned on a 9.4T Bruker MRI scanner, where 1114 images were used for training and 20 were used for testing. Bland-Altman plots, students t-test, and correlation analysis were conducted to compare the generated displacement fields against the gold-standard measurements generated from the tagged images. Results: The generated output displacement fields (Fig 1) showed myocardium shape similar to that in the input images as well as regional bright and dark signal intensities (reflecting different degrees of tissue displacement) at corresponding locations to those in the gold-standard displacement fields generated from the tagged images. Bland-Altman analysis (Fig 2) showed good agreement between measurements from the output and corresponding gold-standard displacement fields, where almost all the measurement differences lied within the ±2SD agreement level. Student t-test showed insignificant differences between all paired measurements (p>0.05). Lin’s concordance correlation coefficients (CCC) were 0.96 & 0.89 for x- and y-displacement fields, respectively. The developed method reduced the time required for generating the displacement field from more than one minute using the InTag software to a fraction of a second. Conclusions: The developed deep learning based method allows for ultrafast and accurate generation of myocardial tissue displacement fields from conventional CMR cine images without the need for acquiring additional tagged images or using special tagging analysis software, which would help reduce scan time and data analysis time and improve CMR value imaging.

Dayeong An, El-Sayed Ibrahim


Headshot of Justin WomackBioModME for Building and Simulating Dynamic Computational Models of Complex Biological Systems

This poster was presented by Justin A. Womack under the direction of Dr. Ranjan Dash

  • Poster (Click to open as PDF)
  • Abstract
  • Authors

Justin Womack Research Poster

Introduction: The progress of biological research has mainly relied on experimental investigations, but is becoming increasingly coupled to computational modeling. Biological systems, their associated regulation mechanisms and disease processes are intrinsically complex, often encompassing a vast number of interacting components that have become too exhaustive to assess experimentally. Computational modeling, the process of simulating complex systems in silico, has emerged as a powerful tool for generating testable hypotheses by predicting different possible outcomes of the biological networks. However, the technical knowledge required to design and solve computational models is a barrier for most experimental and clinical investigators. We have developed BioModME, an R/Shiny web application that simplifies the model building process and consists of all the tools a researcher needs to design and explore their reaction networks.

Materials and Methods: BioModME is designed using the open-source R programming language. The application uses the Shiny and bs4dash packages to build its web framework with bootstrap widgets. We have included the use of ggplot2 and Plotly libraries for data visualization, allowing the user to create high-quality, interactive graphics. The backend ordinary differential equations (ODEs) are solved using the DeSolve package. Source code, tutorials, documentation, and a link to a deployed version of this software can be found at:

Results and Discussion: BioModME provides a model building suite that allows the user to create a reaction network to describe their model. We have used the application to replicate a computational model of cell cycle in the mitosis phase consisting of over 50 reactions and 26 ODEs. The model is built using various chemical and physical laws, including the law of mass action and Michaelis-Menten kinetics for enzymatic reactions. The ODEs are autogenerated and solved in the program backend for the user. The visualization suite provides the user with multiple options to present the model simulation, including pre-built color palettes, line/axis options, and plotting themes. Users can implement changes across multiple instances of a model under different parameter values showing a set of calculated solutions. All model tables, including parameters, equations, initial conditions, and ODEs are exportable in a variety of formats including a well-structured latex report. Users may also download the underlying code for their model in a “ready to run” script in either MATLAB or R.

Conclusion: BioModME provides a web-application resource to help researchers develop mechanistic computational models without the need for complex mathematical derivations or the knowledge of coding/solving mathematical equations using a programming language. The R/Shiny interface allows for a guided model building process creating robust simulations, high-quality graphics, and summary documents.

Justin Womack, Viren Shah, Said Audi, Scott Terhune, Ranjan Dash


Headshot of Shayan ShafieeA machine learning framework for stratifying high vs. low notch-DLL4 expressing host microenvironment for breast cancer bearing subjects

Presented by Shayan Shafiee under the direction of Dr. Amit Joshi

  • Poster (Click to open as PDF)
  • Abstract
  • Authors

Shayan Shafiee Research Poster

Introduction: Nanomedicine and macromolecule drug delivery rely on the enhanced permeation and retention effect in solid tumors, and crosstalk between malignant tumor cells and the non-malignant TME contributes to tumor growth, drug delivery, and therapy efficacy (PMID: 25540894). Delta Like Canonical Notch Ligand4(DLL4) is a protein-coding gene that is responsible for developing blood vessels and plays a role in tumor angiogenesis. With Consomic rat strains differing in inherited levels of DLL4 in host stroma, we have shown that the degree of DLL4-dependent dysfunctional angiogenesis affects tumor growth and metastasis, drug delivery and therapy response (PMID: 32373218). Herein, we propose a machine learning framework to identify and classify hosts with low and high levels of DLL4 expression on tumor endothelium based on the kinetic NIR fluorescence imaging with Indocyanine green dye. Overall objective is to identify tumour bearing animals likely to respond to DLL4 directed therapies.

Materials and Methods: We generated two rat strains, Salt sensitive (SS) with high systemic DLL4 expression and consomic(inbred strains containing a whole chromosome from another strain in their genome) SSBN3 with third chromosome substitution from Brown Norway with low DLL4 expression. We also constructed two novel SSBN3 congenic(inbred strains containing a given genomic region in their genome) xenograft host strains (MV, MW) by introducing segments of BN chromosome 3 into the genetic background of the parental SS strain, such that MV inherited DLL4 locus from SS and MW from BN. A whole body imaging platform for rats was developed, and all four strains bearing identical tumors were imaged. MDA-MB-231 (231), triple negative breast cancer, was used as a Xenograft cell line. Animals were injected intravenously via tail vein with ICG (0.75mg/kg, MP Biomedicals) and simultaneously imaged using a PIMAX4 ICCD camera(Princeton Instruments), equipped with a 25mm lens (Navitar, DO-25, 0.95 f-stop), with an 830nm long pass filter (ThorLabs), excited by 785nm diode laser (~5mW/cm2), was used with a frame rate of 10.6 fps for 6 minutes. Respiratory motion correction was performed with Fourier-Wavelet methods, MATLAB (MathWorks Inc.). Tumor ROIs were automatically drawn on principle component decomposed images using a 2D cross-correlation mapping algorithm. Pixels' intensity was averaged from the ROIs, and a single time series was recorded. Data from tumor bearing SS and SSBN3 rats were used to refine the features and train the classifier. The framework's performance was validated on MV and MW strains (Figure 1. C, L, O). We characterized dynamic perfusion patterns using several features in the NIR time-series. Best pairs of features in terms of performed sensitivity, specificity, and accuracy were selected by a two-step procedure, recursive feature elimination(RFE), and Exhaustive Feature Selection (EFS) by training classifiers for selecting two out of four feature selected in the first step (PMID: 35308965). After feature selection, the classification algorithm becomes a standard binary classification problem, with feature selection as a subproblem. Finally, we implemented an experimental framework to validate the features by combining tumor detection, feature extraction, and classification algorithms.

Results: The performance of the tumor detection algorithm was visually validated, and all the ROIs passed the inspection. The classifier's accuracy, precision, sensitivity, and specificity over the training run in classifying MW and MV as belonging to their respective DLL4 parent strain were 0.84, 0.98, 0.70, and 0.99, respectively, and the sensitivity and specificity were 91.7 and 95, respectively, over the validation run.

Conclusion: We have demonstrated that whole body non-invasive NIR imaging can discriminate identical tumors based on the differential DLL4 expression in TME using an ML framework with high sensitivity and specificity. If developed for dynamic MR imaging used clinically, similar methods will enable patient stratification for DLL4 targeted therapies.

Shayan Shafiee, Mykhaylo Zayats*, Jonathan P. Epperlein*, Sergiy Zhuk* and Amit Joshi


Headshot of Rachel CutlanAccumulation of Axonal Injury and Neurodegenerative Changes Associated with Repetitive Subconcussive Head Acceleration Exposures in a Biofidelic Preclinical Model

Presented by Rachel Cutlan under the direction of Dr. Brian Stemper

  • Poster (Click to open as PDF)
  • Abstract
  • Authors

Rachel Cutlan Research Poster 2022

Subconcussive head impacts with magnitudes below the concussion threshold are sustained during routine participation in contact sports. Subconcussive head impact exposure (sHIE) can lead to decreased concussion tolerance/higher concussion risk. Identifying blood biomarkers that correlate with sHIE may help identify athletes at risk for concussion. Our laboratory developed a biofidelic preclinical model for sHIE with the number and severity of head acceleration exposures scaled to the rat from data collected in our studies of contact sport athletes. The goal of this experiment was to identify blood-based biomarkers for accumulating damage in the brain associated with sHIE. Sprague-Dawley rats were separated into four groups: high exposure (HE), moderate exposure (ME), single injury (SI), and sham. HE and ME received 30 and 8 low-magnitude head accelerations per day, five days per week for four weeks. SI received one high-magnitude head acceleration, sufficient to produce concussion, at the end of the four weeks. Blood was drawn for HE and ME groups at baseline and the end of each week to track progressive blood serum biomarker changes (neurofilament light, NFL; glial fibrillary acidic protein, GFAP; total tau, t-tau; neuron specific enolase, NSE). Blood was drawn from SI and sham rats at baseline and following single injury/sham. All rats had blood drawn at the terminal timepoint, two weeks post injury. Biomarker levels at all timepoints were normalized to baseline. HE NFL levels were significantly elevated relative to baseline (p< 0.05) at all four timepoints during sHIE. The terminal measurement remained significantly elevated, but also significantly decreased from week 4 during the two-week absence of sHIE. ME rats had no significant elevation of NFL during the sHIE period, but the terminal measurement was insignificantly increased from baseline. SI rats saw a significant increase in NFL immediately following exposure. Sham rats saw a significant increase from baseline to terminal, which might partially explain elevated NFL levels at the terminal timepoint. HE, ME, and SI saw significantly higher NFL levels at peak sHIE compared to sham. T-tau and NSE levels also varied significantly between timepoints. HE and ME terminal t-tau levels were significantly increased from all previous blood draws. HE and SI terminal NSE levels were also significantly increased from all previous measurements. No significant changes in GFAP levels were found at any timepoint. This experiment showed a dose-dependent relationship between sHIE and certain blood biomarkers, which may be useful in tracking potential accumulating damage associated with sHIE. The HE group had significant axonal damage indicated by elevated NFL concentration, but not the ME group, showing that limiting excessive sHIE may be protective. The trend of t-tau and NSE peaking after sHIE ended may imply a chronic phenomenon, beyond the scope of this study, that deserves future attention.

Rachel Cutlan, Jack Seifert, Alok Shah*, Rachel Chiariello*, Matthew D. Budde*, Christopher M. Olsen*, Michael A. McCrea*, Brian D. Stemper


Headshot of Ricardo VegaProbabilistic Tractography Improves Measurements of Structurofunctional Connectivity

This poster was presented by Ricardo Vega under the direction of Dr. Brian Schmit

  • Poster (Click to open as PDF)
  • Abstract
  • Authors

Ricardo Vega Research Poster 2022

Probabilistic StructuroFunctional Connectivity (pSFC) was used to characterize changes in brain networks caused by stroke lesions. The base SFC algorithm constrains traditional functional connectivity measures (based on blood oxygen level dependent functional magnetic resonance imaging) by the structural connectivity obtained from diffusion tractography. Diffusion tractography models the structural connections between regions of the brain using magnetic resonance diffusion imaging data. A prior iteration of this used a deterministic modeling approach to obtain the structural connectivity, which is straightforward to model but has substantial limitations, especially for subcortical structures. In this study, we implemented a probabilistic tractography model, which overcomes many of the limitations of deterministic modeling, but is more computationally intense. By utilizing a probabilistic method for modeling white matter (WM) connections, key structural connections to subcortical regions integral for motor control and sensory processing were characterized. Due to the complex structure of the WM tracts that connect these regions, conventional deterministic tractography approaches failed to accurately model these connections. Additionally, segmentation of high-resolution anatomical T1 images was performed to create a 5-tissue type mask and registered to native diffusion space. This mask was the used to calculate diffusion coefficients specific for each tissue type, assisting in accurate tract reconstruction. The tissue types consist of white and grey matter, subcortical grey matter, cerebral spinal fluid, and pathological tissue (i.e. lesions). Our pSFC results demonstrated an overall decrease in brain connectivity in people with stroke compared to controls. Additionally, pSFC had a greater sensitivity to residual connectivity in stroke participants compared to deterministic structurofunctional connectivity.

Ricardo Vega, Brian Schmit


Headshot of Mir Hadi Razeghi KondelajiNIR-II window imaging to assess the impact of inherited Notch-DII4 expression on pulmonary radiation injury

This poster was presented by Mir Hadi Razeghi Kondelaji under the direction of Dr. Amit Joshi

  • Poster (Click to open as PDF)
  • Abstract
  • Authors

Mir Hadi Razeghi Kondelaji research poster 2022

Introduction. Ionizing radiation can cause lung injury by increasing vascular permeability. Recently we reported a physiologically based pharmacokinetic (PBPK) model based on non-invasive dynamic NIR imaging to estimate vascular function and permeability in lungs [PMID: 28567545]. However, the role of inherited genes in pulmonary radiation injury and recovery has not been fully studied. We report the influence of inherited notch-DLL4 expression levels in endothelial cells on radiation injury via NIR-2nd window dynamic imaging of ICG dye kinetics in consomic rat models. Consomic strains were generated by the substitution of the brown Norway (BN) rat (DLL4-low) chromosome 3 into the salt sensitive (SS, DLL4-high) background (referred to as SS-BN3 consomic). We demonstrate the divergent effect of DLL4 expression on radiation induced vascular permeability, which further led to mortality differences, where SS rats (DLL4-high) exhibited high pulmonary mortality compared to SS.BN3 rats following whole thoracic radiation.

Method. SS and SSBN3 rats were bred to differ in the inheritance of 3rd chromosome (Figure 1- A). Animals in each group were divided into two groups, where one group was exposed to 13.5 Gy leg out partial body irradiation. NIR-2nd window dynamic fluorescence imaging of ICG dye uptake and clearance was conducted with a NIRVANA camera (15 minutes, 10.6 frame/second, and 808 nm excitation) at 42 days and 90 days post irradiation. Time dependent images were analyzed by principal component analysis (PCA) to detect lung ROI and time courses across groups were compared (Figure 1-B, and C). PBPK compartmental model parameters were estimated to quantify the permeability- surface area product (PS). Figure 1-D. To evaluate the endothelial cellularity, and the sensitivity of cells to the irradiation, and to evaluate the role of DLL4 expression on the irradiation response, flow cytometry was performed for selected animals from each group. Also, about ten million lung cells per animal were used for CD31+ enrichment to assess the effect of irradiation injury on the CD31+ count in radiated and control groups. A 2- way ANOVA analysis followed by post hoc comparison and performed for statistical evaluations and significance measurements. The produced images and the estimation of PBPK parameter were performed in MATLAB 2021b and statistical significance tests performed with R.

Results. 42 days post injury, the PS value increased to 6.85 ± 1.89 [CL: 2.76- 10.94] mL/min for SS rats after 13 Gy irradiation, compared to 2.36 ± 0.20 [CL: 1.94- 2.78] mL/min for nonirradiated SS rats (p-value<0.05). While, the PS for SSBN3 rats was 2.56 ± 0.21 [CL: 2.11- 3.01] mL/min, and 3.50 ± 0.69 [CL: 2.06- 4.94] mL/min for control and radiated groups(Figure 1-E). This suggests SS.BN3 pulmonary vasculature was resistant to radiation injury. We examined the endothelial cellularity (EC) in lung by flow cytometry (Figure 1-H). In SS rats EC reduced from 9.87% in 0Gy to 5.65% in 13Gy. However, BN3 rats showed a nonsignificant reduction in EC from 7.08% in 0Gy to 5.46% in 13Gy, suggesting radioprotective effects in consomic SSBN3 rats (Figure 1-F). There was significant decrease in Dll4 mRNA in SS rats upon radiation compared to that in SSBN3 rats. Notably, in non-radiated groups, SS rats have higher EC counts and Dll4 mRNA expression than SSBN3 rats (Figure 1-G) suggesting higher DLL4 expression increases radiation injury. Also, the morbidity analysis of animals shows higher rate for irradiated SS rats compared to non- irradiated SS and radiated SSBN3 groups (Figure 1-H).

Conclusion. Our study suggests a significant dependence of radiation injury to the inherited notch-DLL4 expression, which can be imaged in 200g rats with NIR 2nd window imaging.

Mir Hadi Razeghi Kondelaji, Guru Prasad Sharma*, Jaidip Jagtap*, Shayan Shafiee, Christopher Hansen, Tracy Gasperetti*, Anne Frei*, Dana Veley*, Jayashree Narayanan*, Brian L. Fish*, Abdul K. Parchur*, El-Sayed H. Ibrahim, Meetha Medhora*, Heather A. Himburg, Amit Joshi


Headshot of Rachel (Rocky) MazorowActivation Response Profiles for Eccentric Rotating Mass Vibration Motors Used for Sensory Augmentation

This poster was presented by Rachel Mazorow under the direction of Dr. Robert Scheidt

  • Poster (Click to open as PDF)
  • Abstract
  • Authors

Rachel Mazorow research poster 2022

Introduction: Decreased sensations of limb position and movement are experienced by ~50% of stroke survivors, leading to a decrease in upper extremity motor control and difficulty performing common daily tasks1. Although visual feedback of the arm and hand can partly compensate, resulting movements are typically slow and jerky. Several recent studies have explored an alternate compensatory approach: using supplemental vibrotactile guidance (SVG) to enhance closed-loop control of the arm2. Those studies used low-cost, eccentric rotating mass (ERM) vibration motors to provide continuous, graded feedback of limb posture and movement via modulation of vibration intensity. Here, we aimed to compare the stimulus response characteristics of two types of ERM vibration motors used in recent studies of SVG.
Materials and Methods: A multi-axis load cell (85M35A-I40-A-200N12; JR3 Inc) measured mechanical vibrations from two ERM motors driven by a pulse-width modulated signal that ranged from 0% to 100% full scale activation in 10% increments. The motors were a miniature coin-style motor (310-117; Precision Microdrives Ltd.) and a larger pager-style motor (308-102; Precision Microdrives Ltd). Each ERM motor was tested 5 times with a stepped-triangular activation profile (5 s per step). The load cell data were notch-filtered at 60 Hz. A short-time Fourier transform was applied using a Hamming window (128 ms window; 50% overlap). The across-trial mean peak power and its associated frequency were identified at each activation level for each motor.
Results and Discussion: Both motors required ~20% activation to overcome initial stiction. Both motors exhibited similarly shaped vibration frequency curves, with the pager-style motor having a higher peak frequency at full activation. The pager-style motor also had a higher magnitude of peak signal power but exhibited non-linearities in its activation response profile and frequency-intensity curve above 200 Hz. The response profile of the coin-style motor was more linear in both the ascending and descending phases of activation.
Conclusions: The coin-style motor had linear activation response characteristics for both vibration frequency and intensity, and thus may be preferred in applications providing graded SVG of movement variables such as hand position or joint angle. The pager-style motor produced vibrations ~2x as intense as the coin-style motor, but exhibited a non-monotonic amplitude-frequency characteristic that has the potential to make interpretation of the vibrotactile cues more difficult; however, the stronger intensity may be useful for individuals with more significant impairment of tactile sensation. Ultimately we expect these results will prove useful for the design of systems providing SVG of ongoing limb movements.
Acknowledgment: National Institute of Child Health and Human Development grant R15HD093086.
References: 1Rand D. PLOS ONE. 2018., 2Risi N, et al. Journal of Neurophysiology. 2019.

Rachel N. Mazorow & Robert A. Scheidt


Every year, MCW's School of Graduate Studies and the Office of Postdoctoral Education's annual Research Poster Session evaluates graduate students and postdoctoral fellows on the quality of their research, poster design presentation. Winners are selected from both graduate student and postdoctoral fellow entrants.

The winners of this year’s graduate school poster session included two Joint Department students, Dayeong An for "Myocardial displacement field generation from CMR cine images by image-to-image translation deep learning networks," and Rachel Cutlan for "Accumulation of axonal injury and neurodegenerative changes associated with repetitive subconcussive head acceleration exposures in a biofidelic preclinical model"


Learn more about BME Research