Category: Presentation

Saga Helgadottir will give a presentation at the Gothenburg Science Festival 2020.

The International Science Festival Gothenburg is one of Europe’s leading popular science events. Its first edition dates back to 1997, and it is held every year in spring.
This year the festival will take place during autumn, 28 September-4 October. Due to the current situation the festival will be a digital event. The digital festival will be available during the week of the festival.

The contribution of Saga Helgadottir will be presented according to the following schedule:

Saga Helgadottir
Deep Learning for Object Recognition
Deep Learning is a machine learning technique that teaches computers to do what comes naturally to humans: learn by example. In this talk, I will show how Deep Learning can be used to identify objects in images, in particular microscopic particles.

Date: 2 October 2020
Time: 18:08
Duration: 17′
Link: Deep Learning for Object Recognition

Links:
Vetenskapsfestivalen Göteborg (in Swedish)
The International Science Festival Gothenburg (in English)
Full Program

Digital video microscopy with deep learning
Saga Helgadottir
(Invited paper)

Microscopic particle tracking has had a long history of providing insight and breakthroughs within the physical and biological sciences, starting with Jean Perrin proved the existens of atoms in 1910 by projecting images of microscopic colloidal particles onto a sheet of paper and manually tracking their displacements. From the start of digital video microscopy over 20 years ago, automated single particle tracking algorithms have followed a similar pattern: pre-processing of the image to reduce noise, segmentation of the image to identify the features of interest, refining of these feature coordinates to sub-pixel accuracy and linking of the feature coordinates over several images to construct particle trajectories. By fine-tuning several user-defined parameters, these methods can be highly successful at tracking a well-defined kind of particle in good imaging conditions. However, their performance degrades severely at unsteady imaging conditions.
To overcome the limitations of traditional algorithmic approaches, data-driven methods using deep learning have been introduced. Deep-learning algorithms based on convolutional neural networks have been shown to accurately localize holographic colloidal particles and fluorescent biological objects. We have recently developed DeepTrack, a software package based on a convolutional neural network that outperforms algorithmic approaches in tracking colloidal particles as well as non spherical biological objects, especially in the presence of noise and under poor illumination conditions.
In this talk I will give an overview of the history of particle tracking before explaining the details of our solution DeepTrack and finally give an outlook on the field of deep learning in microscopy.

Time and place: Presentation published online on 24 August 2020
SPIE Link: here.

DeepTrack: A comprehensive deep learning framework for digital microscopy
Benjamin Midtvedt, Saga Helgadottir, Aykut Argun, Daniel Midtvedt, Giovanni Volpe
Click here to see the slides.

Despite the rapid advancement of deep learningmethods for image analysis, they remain under-utilized for the analysis of digital microscopy images. State of the artmethods require expertise in deep learning to implement, disconnecting the development of new methods from end-users. The packages that are available are typically highly specialized, diicult to reappropriate and almost impossible to interface with other methods. Finally, it is prohibitively difficult to procure representative datasets with corresponding labels. DeepTrack is a deep learning framework targeting optical microscopy, designed to account for each of these issues. Firstly, it is packaged with an easy-to-use graphical user interface, solving standard microscopy problems with no required programming experience. Secondly, it provides a comprehensive programming API for creating representative synthetic data, designed to exactly suit the problem. DeepTrack images samples of refractive index or flourophore distributions using physical simulations of customizable optical systems. To accurately represent the data to be analyzed, DeepTrack supports arbitrary optical aberration and experimental noise. Thirdly, many standard deep learning methods are packaged with DeepTrack, including architectures such as U-NET, and regularization techniques such as augmentations. Finally, the framework is fully modular and easily extendable to implement new methods, providing both longevity and a centralized foundation to deploy new deep learning solutions. To demonstrate the versatility of the framework,we show a few typical use-cases, including cell-counting in dense biological samples, extracting 3-dimensional tracks from 2-dimensional videos, and distinguishing and tracking microorganisms in bright-field videos.

Poster Session
Time: June 22nd 2020
Place: Twitter and virtual reality

POM Conference
Link: POM
Time: June 25th 2020
Place: Online

Poster Slides