To study how axon growth is affected by the local environment biologists perform extensive experiments, watching the axons develop on different substrates. As axons grow from the neuron cell body they form tree-like structures, with branches forming and withering as they explore their surroundings. In this project, we designed a system which can track individual axons as they grow and branch over time, enabling quantitative evaluation of different aspects of the axon behaviour. The system includes a novel segmentation network that can be aware of not only intensity but also properties such as feature direction and scale. Such properties can be important when analysing images containing curvilinear structures such as vessels or fibres. We propose the General Multi-Angle Scale Convolution (G-MASC) network, whose kernels are arbitrarily rotatable and also fully differentiable. The model man- ages its directional detectors in sets, and supervises a setâÃÂÃÂs rotational symmetry with a novel rotation penalty called PoRE. The algorithm works on pyramid representations to enable scale search. Direction and scale can be extracted from the output maps, encoded and analysed separately. Tests were conducted on three public datasets and the axon samples. Good performance is observed while the model requires 1% or fewer parameters compared to other approaches such as U-Net.
- Tiny model
- Neural networks
- MASC
- Multi-angle and scale convolution
- Rotational invariant
- Rotatable model
- Graph Neural networks
- Convolutional neural networks
- Computation efficiency
- Image segmentation
- Axon
A System for Anlaysing Axon Activity Using Multi-Angle and Scale Convlutional Networks
Liu, Z. (Author). 31 Dec 2022
Student thesis: Phd