Fluorescence microscopy is one of the most common techniques used in cell biology research. It allows researchers to specifically label structures of interest and observe these within living samples. However, despite the popularity of fluorescence microscopy, there is still very little robust guidance on what makes a ‘good’ image. Obtaining good images is a particular challenge when imaging live cells, as the microscope settings required to produce a visually pleasing image can in fact damage (or even kill!) the sample.
Our group aims to reveal what factors impact the quality of images obtained using fluorescence microscopy, how this affects the accuracy of measurements that can be made from images, and how we can develop image analysis methods to improve the quality of microscopy data.
Projects

Fluorescence microscopy image database
In order to assess what image features most strongly correlate with accurate measurements, we aim to create a database of fluorescence microscopy images acquired with a range of different settings. The model organism for this database will be the fission yeast species S. pombe. The growth and size regulation of S. pombe is tightly controlled, and so measurements of features such as the shape and size of labelled intracellular structures should fall within a well-defined range. If measurements made from an image fall outside this range, then this is an indication that the image was not high-enough quality.

Deriving quality metrics from image formation models
Regardless of the sample being imaged and the specific fluorescent labels and microscope hardware being used, all fluorescence microscopy images are formed in the same fundamental way. The stages in this image formation process are mathematically well-defined, and we aim to use this knowledge to investigate whether the quality of a fluorescence microscopy image can be calculated from the statistical properties of the image itself.

Quality-driven image processing
Deep learning-based image restoration is a quickly growing field, allowing users to train networks that can perform tasks such as reducing noise and improving resolution of microscopy images. However, most of these methods are trained with generic image quality metrics such as per-pixel error and structural similarity that are agnostic to the unique properties of fluorescence microscopy data. We aim to instead construct and train networks with the objective of improving microscopy-specific image quality and biological information content.
Publications
Awards
Royal Society University Research Fellowship
‘Biology-driven image analysis for light microscopy’
October 2021 – September 2026
Projects

Fluorescence microscopy image database
In order to assess what image features most strongly correlate with accurate measurements, we aim to create a database of fluorescence microscopy images acquired with a range of different settings. The model organism for this database will be the fission yeast species S. pombe. The growth and size regulation of S. pombe is tightly controlled, and so measurements of features such as the shape and size of labelled intracellular structures should fall within a well-defined range. If measurements made from an image fall outside this range, then this is an indication that the image was not high-enough quality.

Deriving quality metrics from image formation models
Regardless of the sample being imaged and the specific fluorescent labels and microscope hardware being used, all fluorescence microscopy images are formed in the same fundamental way. The stages in this image formation process are mathematically well-defined, and we aim to use this knowledge to investigate whether the quality of a fluorescence microscopy image can be calculated from the statistical properties of the image itself.

Quality-driven image processing
Deep learning-based image restoration is a quickly growing field, allowing users to train networks that can perform tasks such as reducing noise and improving resolution of microscopy images. However, most of these methods are trained with generic image quality metrics such as per-pixel error and structural similarity that are agnostic to the unique properties of fluorescence microscopy data. We aim to instead construct and train networks with the objective of improving microscopy-specific image quality and biological information content.
Publications
Awards
Royal Society University Research Fellowship
‘Biology-driven image analysis for light microscopy’
October 2021 – September 2026