3 PhD positions on the “SEARCH - Searching the perfect one” project (Deep learning for microfluidic...

Updated: 2 months ago
Job Type: Temporary
Deadline: 01 Aug 2020

Project and challenges

Often cell populations are analysed by looking at average values of a population, such as size, shape and expression of proteins. However, a cell population is not homogeneous; it consists of different individual cells with different responses to stimuli.

Microfluidic technologies are a perfect tool to analyse the properties of individual cells. For example, the beating of individual spermatozoa can be measured by using electrical impedance measurements or optical analysis. More valuable information is present in the data, but this is currently not used.

The project will improve single cell analysis on a microfluidic scale, with robust image processing and machine learning techniques. The aim is to generate more knowledge about the

heterogeneity of cells in a population and their individual response on different stimuli, but also to make ground to choose the best cell of the population.

Two application areas will be explored:

(1) crop improvement in collaboration with the company Keygene

(2) spermatozoa analysis in collaboration with the clinics Radboud UMC .

PhD projects

We offer three PhD positions, one for each of the work-packages of the project.

WP1: 1 PhD position on Developing microfluidic chips for single cell trapping

One of the advantages of microfluidics is that the dimensions are on the same scale of cells, making trapping of individual cells possible. For both of our applications, trapping is necessary, but they differ. For the agricultural application, protoplasts from different origins (e.g. upper/lower epidermis, paranchemy, stromata) with different dimensions and properties need to be trapped in a controllable and reliable manner, without harming these vulnerable cells. After that, predictable and reliable gradients of different chemical compounds need to be made, to test the dynamic response of different stimuli on each cell.

In the medical case, you will work on trapping boar spermatozoa. The challenge for this case is to make it applicable for human spermatozoa and translate it to the clinical setting, meaning that every technician can use it (operator independent) in an easy way without manual control or steering. Live cell imaging of the trapped cells will be used for WP2 and WP3.

Contact person:
Prof. dr. Loes Segerink (Electrical Engineering)


WP2: 1 PhD position on Deep manifold learning of cell morphology and motion

To search for the perfect one we need a topological map (manifold) of cells created by deep learning (DL). Such a map can encode an informative, understandable and discriminable feature representation of cells.

The challenge is to extract informative cell morphology like shape, structure, form or size (WP3), and informative cell dynamics like velocity, acceleration, rotation, deformation or growth from dynamic imaging data and combine it into a unified higher‐order feature representation. Deep manifold learning (non‐Euclidean, graph‐based geometric DL methods) will be at the core of this work-package. The geometry of the manifold representation will allow us to interpolate between advanced “cell barycentres”, and enable new ways for controlling dynamic cell cultures via chemicals towards desired properties.

You will use live cell imaging data (WP1), develop robust segmentation together with WP3, and extract cell dynamics via learning networks for optical flow.

Contact person:

Prof. dr. Christoph Brune (Applied Mathematics)


WP3: 1 PhD position on Robust image processing with deep learning

Qualitative morphological differences of cells can be observed across different flasks or experiments conducted in different days. These variations harm generalization of deep networks and their use for automatic assessment or measurement of cell characteristics in real‐time.

You will research image processing and machine learning techniques to train robust‐by‐design models, by addressing the intrinsic challenges of live cell imaging data (from WP1): weakly and noisy labels, and variations in the imaging data due to different recording conditions.

You will investigate new architectural elements and self‐supervised training strategies for learning of robust segmentation and morphological features in deep networks, and support the extraction of robust velocity, acceleration, rotation, deformation or growth features in WP3.

Contact person:

Dr. Nicola Strisciuglio (Computer Science)


View or Apply

Similar Positions