EN
Passive propagation of electric fields can induce ap‑ parent coherence in local field potentials (LFP) record‑ ed over distances of several millimetres hindering their analysis. This issue can be overcome with current source density analysis (CSD). Mathematically, CSD reconstruc‑ tion is an ill‑posed problem which means that many dif‑ ferent possible current source distributions fit the mea‑ sured LFP and the challenge is to find the most probable one. Furthermore, LFP recordings are always noisy, par‑ ticularly in data obtained from freely moving animals, which may affect CSD estimation. Previously, we pro‑ posed the kernel CSD (kCSD) method for reconstruction of the spatial distribution of sources and sinks in biologi‑ cal tissue from noisy data. Here we show how the method parameters can be estimated quickly and reliably using an L‑curve approach. We demonstrated the feasibility of this approach on model data and illustrated its power in the analysis of LFP recordings from linear probes im‑ planted in the olfactory bulb (OB) of freely moving rats. We focused on ketamine‑induced high frequency oscilla‑ tions (HFO, 120‑200 Hz) since, to date, the locus of gen‑ eration of HFO remains unclear. kCSD is a model‑based CSD estimation method which assumes a flexible model of CSD and estimates its parameters from data. L‑curve is a technique for finding the optimal way of weighting the complexity of the model against the difference between model predictions and the actual set of measurements. The LFPs we analysed were recorded from freely moving rats implanted with a 32‑channel linear probe targeted to the OB. Recordings were made at baseline and post in‑ jection of 25 mg/kg ketamine (i.p.). To examine the faith‑ fulness of kCSD reconstruction we tested this method on model LFPs from ground truth data. We showed that the L‑curve provides reliable and practical estimation of regularization parameters for robust kCSD estimation of sources from noisy LFPs. After validating this method, we estimated the current sources from recordings in the rat OB. We found HFO dipoles close to the mitral layer, whereas above it there was little evidence of any phase reversal. kCSD with L‑curve is a robust method for esti‑ mation of current sources from noisy data. It facilitates localization of the sources of abnormal HFO activity to a specific layer within olfactory bulb which is consistent with histology.