Transfer Learning for precise detection of individual cells in multimodal microscopic image data

Automatic segmentation and detection of cells is a fundamental task in relevant medical fields such as histopathology, hematology, and cytopathology. Deep Learning methods show promising results, but often require excessive amounts of data, which is a major barrier to entry, especially for experimental cellular imaging data.
This project aims to develop a state-of-the-art cell detection framework that is applicable or transferable to many different data modalities, i.e. imaging techniques or biological staining protocols, with minimal effort. To achieve this goal, a potent Deep Learning architecture is combined with a newly compiled dataset, consisting of several existing cell-datasets, as well as newly generated synthetic datasets. The latter may be created with algorithmic approaches and generative Deep Learning models. Moreover, the benefit of such a dataset and applicability of Transfer Learning to trained Deep Learning models will be studied. The created framework will be evaluated against state-of-the-art methods and other datasets.

Faculty Supervisor:

Alan Evans

Student:

Partner:

Forschungszentrum Jülich

Discipline:

Computer science

Sector:

Artificial Intelligence; Health and Related Sciences & Technology

University:

McGill University

Program: