New research could hold the key to understanding cell behaviour
Research Fellow Weidi Xie, together with Professors Alison Noble and Andrew Zisserman, has won the Taylor and Francis Best Paper Award for his DPhil research into microscopy cell counting and detection.
The findings of his paper, 'Microscopy Cell Counting and Detection with Fully Convolutional Regression Networks' could mean that the current time-consuming method of manually counting cells becomes a thing of the past, and that previously unattainable data from overlapping or clumped cells is now made available.
This breakthrough brings benefits to a range of medical and biological procedures where cell counting from images is used – for example, to infer a patient's health from their white and red blood cell counts.
The process of cell counting is accelerated by means of an algorithm that estimates the number of cells present, removing the need for researchers to manually count from an image. In the study, 294 cells were detected using the new method, verses 297 detected when counting manually.
FCRN-A applied on Plasma Cells: Only greyscale image is used.
Weidi’s research identified two further benefits of this new technique. Firstly, that it is capable of detecting/localizing cells – for example, cancerous cells. Secondly, that once cancerous cells were discovered, their behaviour could be monitored by separating out each layer for individual analysis. This is especially useful where the cell structure is complex, such as a cell that evolves or adapts, and could help researchers to more fully understand cell behaviour.
Weidi said: “I’m happy to be recognized by the scientific community, and I’m looking forward to contributing more work at the intersection of healthcare, vision and machine learning.
“My current research is in designing self-supervised learning algorithms that allow the system to learn from massive amounts of video sources, and I look forward to further developing this field.”
You can find out more about the people behind the research below: