Listen to world-leading experts in imaging

Fluorescence microscopy is a central technique for quantifying the spatiotemporal distribution of cellular compartments. However, the ability to accurately retrieve biological information from images depends strongly on the quality of the data. The perceived quality of fluorescence microscopy images can be improved by imaging with higher illumination doses, however this comes at the cost of potentially damaging and altering the behaviour of living samples. To reduce the impact of noise when acquiring at lower illumination densities, image processing techniques have been developed, many of which are based on AI. Here, we interrogate quality metrics which are commonly used to assess the performance of image processing methods. These methods are commonly used in microscopy, but have their origins in computer vision. We show that these metrics report not just image quality, but also other characteristic features of fluorescence microscopy images, which can confound their interpretation. We also quantify how predictive these metrics are of successful downstream image analysis for a variety of common tasks, and discuss how and when these metrics should be best used.

‍

Bio: Siân Culley is a Royal Society University Research Fellow who started her group at King’s College London in 2021. Her interests are in how microscopy methods can be tailored to best fit biological questions, with a particular focus on assessing data quality and reliability. Prior to starting her group she did postdoctoral work with Ricardo Henriques at UCL, developing open source methods for super-resolution fluorescence microscopy.