Abstract

The past decades have seen a large democratisation of image analysis software either commercial or freely distributed, allowing segmentation of objects of interest. However, as the dimensions and the resolution of images increase with the new microscopy techniques, scientists’ demands would require more powerful and flexible tools to extract data from features of interest.

In the case of biological studies, it is necessary to compare the assessed specimen to a negative control and then show statistical relevance of the difference. So far, image analysis tools are successful in segmenting simple objects, i.e. oval shapes, filaments, small dots, etc., in a neutral surrounding. Still, the main issues related to the imaging of biological samples, i.e. noise, non-specific background or heterogeneous signal, often hinder the detection of the features of interest. Improving strategies in de-noising, shape modelling or machine-learning could for instance help achieving more precise measurements, giving a better answer to the biological question.

The aim of this presentation is to briefly introduce some examples of image analysis studies conducted at the CRUK Cambridge Institute, stressing the respective successes and difficulties. To a broader extent, we will discuss about the way biologists, mathematicians and computer scientists could better link their respective skills and knowledge.