Microfluidics and Deep Learning for Male Fertility
Abstract
Infertility rates are on the rise globally, and male fertility is in decline. The assessment of male fertility and the selection of sperm for artificial reproduction are current challenges that are well suited to the combination of microfluidics and machine learning. This talk will provide first an overview of our early work in microfluidics for male fertility diagnostics, and microfluidics for sperm preparation. Microfluidic sperm preparation methods simplify and standardize clinical workflow. The next stage in the clinical process is individual sperm selection for intracytoplasmic sperm injection (ICSI, a popular form of in vitro fertilization where an individual sperm is selected and injected directly into an egg). Currently, this selection process remains highly subjective. For this final selection challenge we turn to machine learning and particularly to deep learning. We show how deep learning networks can be trained with user input or DNA data to accelerate, standardize and improve the performance of sperm selection. Specifically, we show how a neural network can teach itself to classify sperm into WHO categories and rank sperm via DNA quality. We use an existing deep learning model, VGG16, pre-trained on the ImageNet database, a collection of >10 million everyday (non-medical) images, and modify the last few layers to either classify or rank cells. For the classification task, we achieve an average true positive rate of 59% and 86%, matching or exceeding past efforts. For the ranking task, we reliably rank sperm within the 77th percentile of DNA integrity. Collectively, our findings demonstrate how microfluidic sperm processing pairing deep learning analysis can enhance clinical selection of sperm.
Biography
David Sinton is a Professor in the Department of Mechanical & Industrial Engineering at the University of Toronto, and the Canada Research Chair in Microfluidics and Energy. He was the Associate Chair of Research in Mechanical & Industrial Engineering, as well as the Interim Vice-Dean of Research in the Faculty of Applied Science & Engineering. Prior to joining the University of Toronto, Dr. Sinton was an Associate Professor and Canada Research Chair at the University of Victoria, and a Visiting Associate Professor at Cornell University. The Sinton Lab is application-driven and develops fluid systems for energy, the environment and fertility. The group developed a library of industrial fluid testing systems now commercialized through the startup Interface Fluidics. Dr. Sinton was an NSERC E.W.R. Steacie Memorial Fellow, and is a FCSME, FASME, FCAE and FAAAS. He serves on the advisory board of the journal Lab on a Chip.
Event Details
- Date:
- 5 February 2020 at 2:00 pm – 3:00 pm
Description
Abstract
Infertility rates are on the rise globally, and male fertility is in decline. The assessment of male fertility and the selection of sperm for artificial reproduction are current challenges that are well suited to the combination of microfluidics and machine learning. This talk will provide first an overview of our early work in microfluidics for male fertility diagnostics, and microfluidics for sperm preparation. Microfluidic sperm preparation methods simplify and standardize clinical workflow. The next stage in the clinical process is individual sperm selection for intracytoplasmic sperm injection (ICSI, a popular form of in vitro fertilization where an individual sperm is selected and injected directly into an egg). Currently, this selection process remains highly subjective. For this final selection challenge we turn to machine learning and particularly to deep learning. We show how deep learning networks can be trained with user input or DNA data to accelerate, standardize and improve the performance of sperm selection. Specifically, we show how a neural network can teach itself to classify sperm into WHO categories and rank sperm via DNA quality. We use an existing deep learning model, VGG16, pre-trained on the ImageNet database, a collection of >10 million everyday (non-medical) images, and modify the last few layers to either classify or rank cells. For the classification task, we achieve an average true positive rate of 59% and 86%, matching or exceeding past efforts. For the ranking task, we reliably rank sperm within the 77th percentile of DNA integrity. Collectively, our findings demonstrate how microfluidic sperm processing pairing deep learning analysis can enhance clinical selection of sperm.
Biography
David Sinton is a Professor in the Department of Mechanical & Industrial Engineering at the University of Toronto, and the Canada Research Chair in Microfluidics and Energy. He was the Associate Chair of Research in Mechanical & Industrial Engineering, as well as the Interim Vice-Dean of Research in the Faculty of Applied Science & Engineering. Prior to joining the University of Toronto, Dr. Sinton was an Associate Professor and Canada Research Chair at the University of Victoria, and a Visiting Associate Professor at Cornell University. The Sinton Lab is application-driven and develops fluid systems for energy, the environment and fertility. The group developed a library of industrial fluid testing systems now commercialized through the startup Interface Fluidics. Dr. Sinton was an NSERC E.W.R. Steacie Memorial Fellow, and is a FCSME, FASME, FCAE and FAAAS. He serves on the advisory board of the journal Lab on a Chip.