Andre Telfer

Description

Recently displayed at Haven's Cafe and the Brain and Mental Health artshow 2023.

Description

In both humans and mice, gait analysis has been used to observe dysfunction in the brain. Gait analysis has been used to provide a measure of brain dysfunction in Parkinson's Disease, Huntington's Disease, Stroke, and Multiple Sclerosis. Mouse models of these conditions also observe changes gait. Traditional systems for performing gait analysis in mice often require the mice to be placed in a specific environment. For treadmill based systems This project aims at performing automatic gait analysis using deep learning with fewer setup constraints.

Collaborators

Vanessa Wong, Emmerson Borthwick, Dana Wymark, Delenn Hills

Description

Using an Spiking Neural Network (SNN) with leaky integrate-and-fire dynamics we decode the finger velocity of a primate completing a reaching task from multielectrode array neural recordings.

Motivation

As the number of companies developing brain-computer interfaces continues to grow (e.g., Neuralink and Synchron), it is critical to investigate low latency and energy-efficient neural decoding methods. We wanted to show that spiking neural networks (SNNs), which can be implemented on neuromorphic chips, can decode such signals and fill this gap.

Additionally, SNNs are closer to the underlying neuron biophysics. Therefore, we hoped that by using realistic neural dynamics, we could learn more useful temporal representations in the data compared to traditional decoding or artificial neural network methods which rely on averaging the number of spikes over time.

Links

• DevPost Article

Collaborators

{Michael Stuck, XingYun Wang, Zachary Friedenberger, Emine Topcu-Can} @ University of Ottawa

Description

Many species exhibit facial expressions corressponding to different states of emotion and wellbeing. Mice are one such species that exhibit facial expressions and recent work has highlighted the ability for DeepLearning (/AI) to automatically recognize them in free-moving mice[1]. Models like these could be widely applied to behavioural research in order to improve animal care. Additionally, it potentially offers a new dimension of emotionality data compared to older methods such as the Open Field Test.

Automatic classification of mouse wellbeing based on facial expressions.

Collaborators

Niek Andresen, Dr. Katharina Hohlbaum, Dr. Alfonso Abizaid

Links

BMv1 Well Being Benchmark

Description

A neuroquery-guided review of Autism Spectrum Disorder. Neuroquery is a query tool which trains a model on thousands of pubmed fMRI papers to identify regions of interest in the brain.

This work uses Neuroquery to predict regions of interest in the brain for Autism Spectrum Disorder in a variety of contexts. These results are then quantified using brain atlasses and compared in-depth to a small number of relevant papers.

Collaborators

Yara Mahafza, Mason Irvine, Prof. Argel Aguilar-Valles, Prof. Amedeo D'Angiulli

Description

In Neuroscience, behavioural studies in animals are a key component to understanding how the brain functions to generate complex cognitive, motivational, and emotional processes. Behavioural assays are a cornerstone of this behavioural research, and the Open Field Test (OFT) is a common behavioural assay used to test the anxiolytic effects of drugs, motivation, locomotor activity, fear responses and exploration

Experiments using the OFT generate many hours of video which need to be carefully scored. Manual analysis of these videos is a labour intensive process, in many cases taking weeks, and is prone to reliability/consistency issues. Several commercial automated tools exist for scoring OFT videos. However these tools are very expensive, and may perform worse than more recent Open Source solutions [ref] (for example, they may struggle with wires moving in front of the camera). Furthermore, the Open Field Test is frequently modified and closed-source nature of commercial solution drives many researchers back to manual scoring if their needs are not met in entirety.

This project uses an open source deep learning tool (DeepLabCut) in order to automate scoring of OFT videos. We introduce a video calibration step in order to account for differences between videos (current tools favour a fixed setup which many labs using shared experimental spaces do not have) and more accurately translate between pixels and real distance units. We also introduce an API that allows researchers to extract new features from dozens of hours of video in only a few lines of code.

Collaborators

Dr. John Lewis (University of Ottawa), Dr. Oliver van Kaick (Carleton University), Dr. Alfonso Abizaid (Carleton University)

Links

https://github.com/A-Telfer/bapipe-keypoints

Description

c-Fos expression can occur in neurons after intense activation, and can provide information on cell growth and plasticity [ref]. Currently, manual cell counting is the gold standard. This is time intensive and often takes many weeks, as experiments can produce hundreds of images, each with hundreds of c-Fos expressing cells.

ImageJ is a tool that has been used to semi-automate cell counting. The downside of this approach is that parameters such as pixel intensity thresholds, cell sizes, and cell shapes may need to be adjusted per image. Additionally, images that contain areas of background noise are difficult to extract accurate cell counts from.

Here we explore alternative methods for performing cell counting for c-Fos expressing cells.

Collaborators

Aylar Rahimi, Brenna MacAuley, Andrea Smith, Dr. Alfonso Abizaid

Links

YOLO c-Fos Expression Detection