Identifying and separating out features within an extensive image set requires expertise but is very tedious, often taking months or even years. Although there are processes that will select certain features based on a specific threshold, they don’t always get it right.
Sydney University PhD student, Veronika Valova is studying amyloid plaques in Alzheimer’s disease. She acquired 3D images of amyloid in mouse brains by light sheet microscopy. However, as each sample contained thousands of plaques of variable size, it was impossible to set thresholding levels that identified plaques reliably.
Enter deep learning. This is a type of machine learning that imitates the way humans gain certain types of knowledge. It enables a computer to be trained to do time-consuming and laborious tasks. Ms Valova worked closely with Microscopy Australia’s Sydney Data Specialist to harness deep learning to analyse her data. A model was trained to select plaques based on their shape.
Although this approach, did not give perfect results, it sparked a “close enough–good enough” conversation regarding large data: how much error can be present and still allow meaningful conclusions to be drawn? After considering the results in their biological context and comparing them to other studies, Ms Valova confirmed that the plaque selection was “good enough” to be useful to her research. Her work also demonstrates the issues, value and potential of deep learning to image analysis of microscopy data.
Amyloid plaques (blue: hippocampal, red: cortical) in a mouse brain, identified by deep learning.
March 7, 2023
