Figure 2
Detection, tracking and signal measurement for densely packed single-neuron analysis. (A) 3D nuclei detection method with anisotropy correction, multiscale detection and filtering that produces 3D spots locations. While we show that results are more accurate than the one obtained with available spot detection methods for this task, they still contain spurious detection or regularly miss detecting nuclei along the whole sequence leading tracking software to fail in this context. (B) Detected positions of nuclei in 3D for every time frame of a 3D + time sequence grouped together. Colormap represents time and the bottom image is a zoom in the center of the cloud. (C) The same 3D + time point cloud after rigid and non-rigid registrations forms visible clusters. (D) Clustering of dense region by DBSCAN enables to identify those clusters and remove noise. (E) Final nuclei trajectories in the original spatial coordinates are identified by successive individual detections over time with the same cluster id (here the same color). Merged trajectories and missing detection can easily be reconstructed at this point (see Supplementary Fig. 4) (F). For each detected nucleus, the raw GCaMP signal is measured over time in a volume around the nuclei. 500 individual neurons signal from the same MB cell bodies are displayed here. (G) Normalized signal displayed only for responsive neurons selected as having a peak above 0.1 (see online methods). (H) Count of responsive neurons per fly for a group of 23 naïve flies. (I) Mean intensity of responsive neurons per fly during air and OCT stimulation windows showing a highly significant difference (Wilcoxon signed-rank test, p-value: 7.3089e-68).
