The mouse brain is a highly specialised machine where each component plays a very specific
role. We find organised structures and patterns everywhere, with columnar and layered design
being a hallmark of brain cortex. Within this topographically organised neural networks we find
that their components have well-defined roles. This specificity is reflected on their morphology,
electrophysiology, connectivity pattern, and relative location. In order to fully understand these
networks it is necessary to exploit all information available.
Extracellular recordings are one of the most widely used techniques by neuroscientists,
specially when interested in retrieving information from large populations of neurons with
submillisecond precision. However, spatial information and cell classification are very limited
and existing methods do not take advantage of the information available from high density
electrode arrays recordings.
In this thesis, I introduce two novel models that exploit spatial information conveyed by the
extracellular signal recorded on this type of arrays. A simple model to localise the neural soma
in three dimensional space and a model to parametrise salient morphological features of neurons.
These models provide information that will prove useful when studying detailed organisations in
neural networks. The localisation and morphological of neurons requires models that captures
the features that impact the pattern induced at short recording distances as those achieved on
high density electrode array recordings.
Both localisation and classification must be tested for different morphological classes, as found
in cortex. In order to evaluate the models we must first generate realistic simulated data on which
models could be tested. Localisation must hold irrespective of neuronal type while classification
is dependant on neuronal type. The different simulated neurons should reflect the different
signal patterns seen on real recordings. After validate the models on realistic simulated data we
have to evaluate the models on real recordings.
The localisation algorithm successfully recovers the position of simulated neurons with low
errors for distances within expected ranges in cortex. When localising neurons from real
recordings we obtained a distribution of positions that agree with expected ranges in cortex. We separated simulated morphological classes using our classification model, validating the
model as a tool to identify morphological classes from extracellular recordings. We also tested
this model on real recorded data and using its parameters we identified putative morphological
classes. We then verified these classes had different response properties to stimuli and firing
patterns, supporting our theory that by using the amplitude pattern of the extracellular
potential we can identify different neuronal types.
In this thesis we combined the recovered spatial information with response properties from
neighbouring neurons, to characterise the functional topographic organisation in deep layers of
core auditory cortex. Finding a fractured representation, with local populations having similar
response properties and high signal correlation on average, but large differences in response
properties were possible. In agreement with recent imaging studies from upper layers of mouse
auditory cortex that report smooth tonotopy on a large scale, but fractured tonotopy on a fine
scale.