Classification and analysis method of local field potentials recorded from rat somatosensory cortex

A local field potential (LFP) is a particular class of electrophysiological signals, which is related to the sum of all dendritic synaptic activity within a volume of tissue. This type of signal is recorded using a low impedance extracellular microelectrode, placed sufficiently far from individual local neurons to prevent any particular cell from dominating the electrophysiological signal and reflects the sum of action potentials and slower ionic events. The microelectrode measures the electrical potential difference (in volts) between the tissue and a reference electrode placed relatively near the recording electrode. To extract the LFP form the signal acquired, is used a low-pass filter that removes the spike components allowing only the signal of interest, i.e., field potentials. The LFP represents the accumulative activity of the observed area, whereas, the spike data represents the activity of a particular cell under a given situation. The LFP is composed of the more sustained currents in the tissue, typical of somato-dendritic currents and the major slow current is the postsynaptic potential (PSP). LFPs are very important to understand the brain activity and network so they are widely studied. Usually, due to the big variability in the individual recordings, it is common to study the average signal obtained from the individual recordings. In this way the average signal obtained is very clean but during this averaging operation many useful information present in the individual recordings are lost. Thus, for a deeper understanding of the neuronal network activity it is necessary to analyze directly the individual LFPs (referred as single sweep). This work proposes a new method for the single sweep analysis and classification. The method is composed by four principal steps: pre-processing, processing, waveform recognition and clustering. During the pre-processing the principal points that characterize the single sweep are detected. These points are than used during the processing step to create a single sweep template. To reduce the oscillation inside the template, it is smoothed using well defined smoothing techniques. The next step is the recognition of the defined template in each single sweep (known as template matching). This is done through two techniques: the matched filter method and the contour method. After the single sweeps recognition, the classification of the detected sweeps is performed. Three different clustering techniques are implemented: the K-means, the SOM (self organizing maps) and the hierarchical agglomerative. Through this method the single sweeps are classified into different binds according to their shape. Analyzing the various techniques used, we found that the contour method has demonstrated to be the best one in the single sweeps recognition, while the clustering methods proposed generates almost the same results so it is difficult to specify the best one. To understand the neuronal network activity at a greater level, study of the single sweep LFPs are crucial. Due to the fact that the different shapes in the LFPs represent different network activity, a fine classification method is needed to classify the single sweeps based on their signal shapes. The method proposed in this work performs this classification with a good precision. Though it is difficult to apply many computational operations on a huge dataset as in the case of single sweeps, we tried to use a moderate set of operation that obtained satisfactory results