To understand how networks of neurons process information, it is essential to monitor their activity in living tissue. Information is transmitted between neurons by electrochemical impulses called action potentials or spikes. Calcium-sensitive fluorescent probes, which emit a characteristic pulse of fluorescence in response to a spike, are used to visualise spiking activity. Combined with two-photon microscopy, they enable the spiking activity of thousands of neurons to be monitored simultaneously at single-cell and single-spike resolution. In this thesis, we develop signal processing tools for detecting cells and cellular activity from two-photon calcium imaging data.
Firstly, we present a method to detect the locations of cells within a video. In our framework, an active contour evolves guided by a model-based cost function to identify a cell boundary. We demonstrate that this method, which includes no assumptions about typical cell shape or temporal activity, is able to detect cells with varied properties from real imaging data.
Once the location of a cell has been identified, its spiking activity must be inferred from the fluorescence signal. We present a metric that quantifies the similarity between inferred spikes and the ground truth. The proposed metric assesses the similarity of pulse trains obtained from convolution of the spike trains with a smoothing pulse, whose width is derived from the statistics of the data. We demonstrate that the proposed metric is more sensitive than existing metrics to the temporal and rate precision of inferred spike trains.
Finally, we extend an existing framework for spike inference to accommodate a wider class of fluorescence signals. Our method, which is based on finite rate of innovation theory, exploits the known parametric structure of the signal to infer the unknown spike times. On in vitro imaging data, we demonstrate that the updated algorithm outperforms a state of the art approach.