Doctoral students Brendon Lutnick, center, and Brandon Ginley, right, presented research at the Pathology Visions Conference. At left is mentor Pinaki Sarder, PhD.

Doctoral Students Present Research on Digital Pathology Innovations

Published November 28, 2016 This content is archived.

story by dirk hoffman

Two doctoral students in the Department of Pathology and Anatomical Sciences presented their research on digital pathology innovations at the Pathology Visions 2016 Conference, Oct. 23-25 in San Diego.

“Brendon and Brandon are self-motivated, coming up with solutions to problems with little direction. ”
Assistant professor of pathology and anatomical sciences
Print

Brandon Ginley and Brendon Lutnick work in the lab of Pinaki Sarder, PhD, assistant professor of pathology and anatomical sciences.

New Process to Identify Neutrophil Extracellular Traps

Ginley is pursuing a doctoral degree in computational cell biology, anatomy and pathology. His work is centered around the application of computers to traditional fields of biology, anatomy and pathology.

He presented research that details a method he developed to identify Neutrophil Extracellular Traps (NETs).

Neutrophils are white blood cells that help defend the body from pathogens, mainly by phagocytosis, when a cell physically ingests something else, such as bacteria, and breaks it down inside.

In certain conditions, instead of ingesting a material, a neutrophil may activate a cellular death pathway that causes it to form a NET.

A NET is a highly antimicrobial combination of histones, chromatin and granulocytic components. The formation of a NET causes the neutrophil to die.

Because NETs were only first discovered 10 years ago, there is still a large amount of unknown information. However, it is hypothesized that the use of NETs may prevent bacterial dissemination by coating an infection. 

Computational Methods Aimed at Speeding Research

In Sarder’s lab, Ginley is developing computational software that is able to automatically identify NETs in fluorescent images.

“We have already completed a method that is able to accurately classify NETs versus neutrophils in flow cytometry images,” he says.

“This technique takes advantage of the morphology of NET and non-NET objects. NETs are long and spindly whereas neutrophils are round and circular,” Ginley says.

“These characteristics help us distinguish between the two images digitally and automatically.”

Ginley has also developed a method to identify NETs from in vivo confocal images of mouse lung tissue with aspergillus hyphae fungal infection. This method, still in development, is the major goal of his future work.

“Our aim is to develop computational methods that can greatly speed research into this new field and enhance the ability to support or refute biological hypotheses on the functions of NETs,” he says.

In Vivo Data From Swedish Collaborator Utilized

Ginley credits his collaboration with Brahm H. Segal, MD, professor of medicine and chief of infectious diseases, for making the research possible.

Segal provided all of the data for the flow cytometry experiments and procured in vivo data to work with from a collaborator in Sweden, Constantin Urban.

“Dr. Segal’s biological knowledge is the only thing that allows me to make relevant hypotheses and conclusions on NET data. I am very thankful to him for allowing this research to be possible and mentoring my analysis along the way.”

Labeling Glomeruli to Aid Renal Histopathology

Ginley also presented a poster on his research involving automatic labeling of glomeruli to aid renal histopathology.

“If small cellular structures in your kidneys become damaged, your ability to properly filter blood may be affected,” he says.

“Typically, as renal disease progresses, the total amount of filtered blood decreases while the amount of proteins filtered to your urine increases, resulting in a condition known as proteinuria.”

At this stage, doctors often take punch biopsies — removing a small section of kidney, staining it with a histological stain to bring out cellular compartments and examining it under a microscope.

Seeking Faster and Better-Defined Quantification

Of particular interest to a pathologist is the glomerulus, a tiny ball of capillaries at the start of blood filtration.

“In order to determine if a glomerulus is damaged, a clinician will look at the tissue, find the glomeruli and count by hand certain features of it,” Ginley says.

These features include the total number of cells, the overall volume and the capillary space or luminal space.

“There are many more features that can help to predict disease,” Ginley says. “However, the main issue is that a pathologist must do all of this by hand. Not only is it difficult to visually estimate the sizes of structures — it is incredibly time consuming.

“The lack of firm quantification, in addition to the time required per sample, causes frequent misdiagnosis leading to inefficiency in the health care system and mistreatment of patients.” 

Filtering Differentiates Between Textural Patterns

Within renal pathology, there’s high demand for software that can break down the renal architecture and quantify various glomerular compartments, but none exists because of the complexity of renal tissue.

“We believe that computational solutions for renal pathology are not so futuristic,” Ginley says.

“We are developing software that can provide pathologists with quantified numbers of different glomerular structures so they can make a more informed decision on treatment and diagnosis.”

The goal, using a process called Gabor filtering, is to make a computer able to accurately draw a line around any and all glomeruli, Ginley says.

A Gabor filter is a linear filter used for edge detection. It simultaneously filters information in both the spatial domain and the frequency domain.

“In this way, it acts much like simple photo receptor cells in our eyes,” Ginley says. “Gabor filtering is very good at identifying two different textural patterns from each other.”

Helping Clinicians Make Better Diagnostic Decisions

Ginley hopes the research will result in a tool that helps clinicians make diagnostic decisions.

“We would like for a pathologist to be able to upload a picture of a renal biopsy, click a button and then all the glomeruli are identified and the amount of each structure within each glomerulus is quantified.”

“We believe that these numbers, provided to a clinician, will give a much faster and more accurate read on which disease is present and at which severity. In addition, we believe this will also help in tracking the usefulness of therapies for renal disease.”

Ginley says the research was made possible through the guidance of John E. Tomaszewski, MD, professor and chair of pathology and anatomical sciences, whose “vision, experience and mentorship encourages the growth of digital solutions for modern-era medicinal obstacles.”

Increasing Speed at Which Images Can Be Segmented

Brendon Lutnick is also pursuing a doctoral degree in computational cell biology, anatomy and pathology under Sarder’s mentorship.

He has always had an interest in digital image manipulation and, in his spare time, enjoys taking photographs and making movies. His self-taught skills in these areas, he notes, directly apply to his research.

Lutnick presented a poster of his research involving an unsupervised algorithm that segments an image into biologically relevant compartments.

The algorithm is parameter-free, needing no input to determine the optimal classification. The code works by modeling an image as a connected network where pixels are connected by the color distances between them.

The network is then cut into segments so that the classification minimizes an energy (optimization) function, known as Potts model Hamiltonian, which is used in theoretical physics to model electron spins.

“My contribution to this method is to increase the speed at which images can be segmented,” Lutnick says.

“Using statistical assumptions, I was able to make the method exponentially faster with the ability to work on large images.”

Aiming to Integrate AI Into Biological Data Sets

Originally, segmenting a 256-by-256-pixel image took upwards of 72 hours. Lutnick’s method allows the classification to be completed in as a little as two seconds with the ability to run images with millions of additional pixels.

“This method can be applied to any big data set, intelligently finding groups within it. This includes biological and non-biological images, genomic data or even clustering people in a large data set such as Facebook,” he says.

“All that is needed is a data set with any number of features representative of each data point.”

Lutnick’s next project is to develop supervised neural networks with applications in biological classification problems.

“This is a network that is modeled after the architecture of the brain and is given a training data set to learn the correct classification scheme,” he says.

“Like the human brain, these networks learn to interpret data correctly based on training. This type of artificial intelligence is ever more frequent but has not yet been highly integrated into biological data sets.”

Students Appreciate Mentor’s Patient Guidance

Both students are grateful for the opportunity to learn under Sarder’s mentorship.

“Working with Dr. Sarder is great because it offers me the opportunity to try out my ideas without constraint and with guidance available when needed,” Lutnick says.

“Dr. Sarder is one of the kindest and most patient individuals I have ever met, with a razor-sharp mathematical and biological intelligence and a strong passion to educate young minds,” Ginley adds.

Unique Skill Sets Propel Students’ Research Efforts

Sarder met Ginley and Lutnick as undergraduate students and was immediately impressed by their intellect.

“I have worked with many undergraduate students, and Brendon and Brandon stand apart,” he says. “Both came into my lab with minimal computational skills, and the speed with which they became acclimated was remarkable.”

Sarder says both students possess valuable skill sets.

“Brendon is improving his analytical skills every day and views algorithm development in a unique way that allows him to impressively formulate novel approaches,” he says.

“Brandon, while also gifted at analytical formulations, has a unique skill in writing,” Sarder says. “He is able to explain very complicated theories and formulations using plain language, which allows him to tie his computational work seamlessly into the biological domain.”

The two students complement one another and often work in close proximity, bouncing ideas between themselves, he notes.

“Brendon and Brandon are self-motivated, coming up with solutions to problems with little direction,” Sarder says.

“The fact that they were selected to present work this early in their graduate career is unique and a testament to their work ethic and intelligence.”