Lectures

Access to the 2020-2021 (pandemic recordings) of our machine learning and advanced machine learning (deep learning) courses are available from here. These include videos, slides, jupyter notebooks and solutions.

I’m slowly uploading recordings or lectures and group tutorials to Youtube

Group Tutorials

We run tutorials on a variety of topics including machine learning, medical image processing and foundation neuroscience.

Below are links to slides and notebooks:

Methods for diffusion MRI analysis in recent neonate studies

Capsule networks

Gaussian Processes models in developmental neuroimaging

Using Animal Models in Perinatal Brain Research

Registration with Deep Learning

Software

The following open source code and software packages have been developed by our lab:

dHCP surface to template alignment - scripts to align neonatal cortical surfaces to the neonatal surface templates available here

Developing Human Connectome Project (dHCP) structural pipeline - pipelines for neonatal cortical surface extraction

Geometric deep learning for cortical surfaces - code for benchmarking a range of geometric deep learning methods on neonatal cortical surface data

MSM - a tool for multimodal alignment of cortical surface data. This is the image registration software used by the Human Connectome Project (HCP) pipelines and in the HCP parcellation paper

ST-fMRI - code for spatiotemporal deep learning on functional MRI data

SVRTK - Reconstruction tools for fetal MRI

newMSM - the new, reworked version of Multimodal Surface Matching software

Datasets

Below are freely available imaging atlases and datasets we have curated as part of our research:

Left-right symmetric cortical surface atlas for dHCP - we extend on the work of Bozek et al. (2018) to increase the age range to 28 - 44 weeks’ PMA, and enforce left-right vertex correspondence

Preprocessed data for benchmarking geometric deep learning on neonatal cortical surfaces - sphericalised neonatal cortical surface data in native and template space, used to benchmark a range of geometric deep learning methods on surface segmentation and phenotype prediction