Project overview:

Big data projects for mapping nervous systems are revolutionising neuroscience. The University of Melbourne has made significant investments in next-generation microscopes to acquire ultra-high resolution (nano-scale) image data of nerves, as well as in the computational infrastructure needed to analyse this data, but there’s a missing link. Currently, neuroscientists are limited to manual analyses of small samples of this data because these procedures have not yet been automated and deployed at scale. To stay on the cutting edge, neuroscientists need to adopt scalable computing architectures for data visualisation and analysis.

I collaborated with neuroscientists from Keast & Osborne laboratory led by Professor Janet Keast and Dr Peregrine Osborne to begin powering this transition, by developing automated image processing tools to find and measure the properties of the neurons which make up peripheral nerves, spinal cord and brain from high resolution (nano-scale) electron microscope images.

I worked closely with the lead investigator Dr. Calvin Eiber to develop a full stack deep learning and image processing pipeline for analysing electron microscopy images of nerves that could automatically detect and trace the profiles of myelinated and unmyelinated axons.

The morphological properties of myelinated and unmyelinated axons are critical for accurately simulating responses of nerves to develop bioelectronic medical implants and understanding neurological conditions. Undertaking these measurements currently requires laborious manual annotation.

Through this project, I have provided Keast & Osborne laboratory researchers with an intensive, hands on experience in using machine learning for microscopy image analysis based on UoM cloud and HPC resources. The researchers now have a working pipeline implemented on this infrastructure that is derived from a published protocol and has been validated using open-source data shared by the authors. The pipeline is already in use locally as a functional semi-automated tool for segmenting myelinated axons, significantly accelerating the rate at which annotation of new data can be accomplished. Although the pipeline offers high level of accuracy of processing unmyelinated axons at the pixel level, further development needed to achieve segmentation of individual unmyelinated axons, which are closely packed within peripheral nerves. Our collaborators are now well positioned to further develop the pipeline to achieve high level of accuracy for the unmyelinated axon segmentation.

Project team

Zaher Joukhadar, Dr Calvin Eiber, Prof Janet Keast, Dr Adam Blanch, Dr Peregrine Osborne, Dr Kristal Spreadborough, Thanaboon Muangwong, and Dr Mar Quiroga

Technologies:

• Python
• Image Processing
• Deep Learning