Researchers at Indian Institute of Science (IISc.) have developed a new graphic processing unit (GPU) based machine learning algorithm called Regularised, Accelerated, Linear Fascicle Evaluation (ReAl – LiFE), which will help to obtain a better understanding and in the prediction of connectivity between different regions of human brain.
This algorithm can help analyse extensive data generated from diffusion Magnetic Resonance Imaging (dMRI) scans which helps scientists study the connectivity in the brain at a speed, which is 150 times higher than a regular desktop computer or existing state-of-the-art algorithms. The study has been published in the journal Nature Computational Science.
With the study, researchers tried to study the wiring of different parts of the brain which helps in performing various computations.
While these patterns can be studied in animals through invasive techniques, in humans, dMRI is used to infer white matter patterns. Through it, scientists can track the movement of molecules to create a comprehensive map of connectome, which is a network of fibres across the brain.
“Even though it is difficult to pinpoint the connectomes, we are trying to infer information highway network by looking at traffic flow patterns (if molecules are like cars). We look at the movement of water molecules in the brain and we try to infer where the wires are. The water molecules have to travel along the length of the cables (axons), which have connected various parts of the brain. By measuring these lengths of water molecules, we are able to infer which areas are connected,” explained Devarajan Sridharan, Associate Professor at the Centre for Neuroscience (CNS), IISc., and corresponding author of the study.
He further added that this technique requires a lot of computation which can be carried out in an efficient manner through GPUs.
“Tasks that previously took hours to days can be completed within seconds to minutes,” he added.
The accurate identification of information networks, conventional algorithms matched the predicted dMRI signals from the inferred connectome with the observed dMRI signal.
A similar algorithm called LiFE (Linear Fascicle Evaluation) was developed earlier to carry out optimisation, but since it worked on traditional CPUs, the computation was time-consuming.
“In the new study, Mr. Sridharan’s team tweaked their algorithm to cut down the computational effort involved in several ways, including removing redundant connections, thereby improving upon LiFE’s performance significantly. To speed up the algorithm further, the team also redesigned it to work on specialised electronic chips – the kind found in high-end gaming computers – called Graphics Processing Units (GPUs), which helped them analyse data at speeds 100-150 times faster than previous approaches.” said a press release from IISc.
This algorithm will have various applications in the field of health, including disease diagnosis and behavioural studies.
“Understanding brain connectivity is critical to uncovering brain-behaviour relationships at scale,” said Varsha Sreenivasan, PhD student at CNS and first author of the study.
While certain patterns of brain connectivity can explain the inter-individual differences in the attention test scores which help determine behaviours, a previous version of the same algorithm can also help distinguish between Alzheimer’s patients and healthy age matched controls just by measuring brain connectivity.
