Neural networks are a subset of algorithms within the machine learning sector

Neural networks are a subset of algorithms within the machine learning sector which are designed based on the human brain and its biological neural networks with the intention to recognise patterns. According to Feldt (2011) there are three different ways in which a neural network can be described which are functional, anatomical and effective connectivity. The aim of this study was to analyse brain data collected from awake rats and to observe any patterns between neurons and to look for a possibility of any functional connectivity. This refers to a potential relationship between the activities of two neurons but with no indication of how their relationship is actually conciliated. A functional connection between two neurons is only plausible when it can be predicted that one neuron will be fired based on the activity of another neuron (Feldt, 2011).
In the last thirty years, functional and structural neuroimaging studies have progressed the advancement and knowledge of the different regions of the human brain and their functions. Neuroimaging techniques such as functional Magnetic Resonance Imaging are used to examine specific or more general functional connectivity patterns. There are various studies examining the functional connection between different brain regions as the co-activation level of functional MRI time-series (Biswal et al, 1997, Greicius et. al 2003, Lowe et. al 2000). Volunteers were asked to relax and to not focus on anything in particular, during which their level of unprompted brain activity was measured. Biswal et al., (1997) showed that during a rest phase, the left and right hemispheres of the primary motor network displayed a high correlation level between their fMRI BOLD time-series (Biswal et al., 19995, Biswal et al., 1997). This suggested continual information processing and a functional connectivity between both regions when the brain is resting (Biswal et al., 1997, Cordes et al., 2000, Greicius et al., 2003, Lowe et al., 2000).
Advancements within analysis techniques have led to researchers being able to use graph analytical methods. This allows researchers to study the overall structure of the brain with a high amount of perceptual detail. This has led to the discovery that the brain ultimately functions as one integrative network, with all its regions and sub-networks associated with each other. Graph theory acts as a theoretical framework, meaning that questions can be answered such as how functional connectivity within the brain is categorised and how does the brain incorporate information from different sub-sections. This means that the structure and function of specific or more general networks of the brain can be examined in more detail (Bullmore and Sporns, 2009, Sporns et al., 2004, Stam et al., 2009, Stam and Reijneveld, 2007). This also acts as crucial reasons for why this study was carried out.
Brain networks can be defined in graph theory as G = (V,E), with V being the amount of nodes mirroring the brain region and E being the functional connection between the brain regions. Nodes can be represented as cortical regions which can also be large-scale brain regions based on a pre-existing template, for example the Brodmann template or fMRI voxels. Also, the level of functional connectivity between regions is assessed by the level of correlation between the time-series of both regions. This level between each pair of nodes is then computed which leads to a connectivity matrix. Determining if a connection exists between two nodes is examined by whether their level of functional connectivity surpasses a pre-defined threshold. This then leads to a model of the brain being created as a functional network with connections between regions that are linked.
Graph theory is good for studying intricate networks such as biological systems like the human brain. This is because a graph contains key properties such as a clustering co-efficient, characteristic path length, node degree and and centrality (Reijneveld et al., 2007, Sporns et al., 2004, Stam and Reijneveld, 2007). A cluster co-efficient can provide researchers with information such as the neighbour clustering within a graph, which shows how close the neighbours of nodes are connected. Also, the characteristic path length in a graph shows how close a node of the network is connected to every single other node within the network, which gives researchers an insight into how connected the network is on a large-scale and how information can be incorporated into different regions. The degree of a node refers to the number of connections that a node has and can display highly connected hub nodes in the brain network. How hubs form can be examined by using centrality measures, which shows how many of the travel routes within a network go through a certain node of the network. A node is considered to have a high level of centrality if it allows a significant number of the shortest travel routes in the network, meaning that it has an important role in the overall communication of a network. Overall, these graph values are extremely crucial to neuroimaging data because they provide significant information about the structure of a network.
Graph analysis techniques have been used in recent fMRI studies with success. For example, Achard et., al (2006) found that resting fMRI studies showed that functional connectivity is well-organised during rest which supported MEG and EEG studies by Bassett et al., (2006). These studies have proven that the brain is organised according to a small-world network. A small-world network has a high level of local connectivity with a very short travel distance between the nodes within the network. Recent studies from Achard and Bullmore, (2007) have supported these findings, showing that the brain network is well-organised on a regional scale.
The data in this study was analysed by looking for spike patterns within the brain acitivity data using Python. The data used was collected using a “Phase3” Neuropixels electrode array which was inserted into the brain of an awake mouse for around an hour. Neuropixels are extremely beneficial to the progression of neuroscience for multiple reasons. For example, Steinmetz et. al (2018) stated that neuropixels are capable of isolating single neuron activity better than technologies used previously. Also, there are no gaps within their coverage site so the neurons will be reliably recorded. Using a neuropixel on a mouse, such as in this study, meant that multiple brain structures could be recorded at the same time. The neuropixel used in this data set recorded from the visual cortex, the hippocampus and some areas of the thalamus. The analysis of this data is crucial because it allows researchers to discover the ways different brain regions such as the hypothalamus and visual cortex are functionally connected.