Fig 1.
ROI falling under detection threshold and position reconstruction.
The images show the temporal progression (t1-t3) of an ROI from data set 2. Neuron x is visible at the center in t1 and t3, but not detected at t2. Its position is inferred from its nearest neighbors (a-h) in previous frames. The white bar represents 10 μm.
Fig 2.
Maximum projection of a volume recorded from a C. elegans head and result of the segmentation.
Panel A is a maximum projection of a volume centered on a nematode head acquired in the GCaMP channel (data set 1). Panel C reports the result of the detection process of the proposed algorithm. Panels B and D report a zoom-in of the first two panels respectively. White bars represent 10 μm.
Table 1.
See Elegans’ automatic detection performance for 3 data sets.
The table shows the number of correctly (TP), undetected (FN), and incorrectly detected (FP) neurons, their rates, and accuracy for each data set. See Elegans’ performance is higher for data sets 1 and 2 (obtained with 60x confocal microscopy) compared to data set 3 (acquired with light-sheet microscopy). These results are consistent with other detection algorithms and enable automated detection of most neuronal spots, with the ability to manually add missed regions.
Fig 3.
Comparison of detection results.
The figure shows for the same acquisition the results of automatic segmentation in the case of See Elegans (panel A) and two other tracking algorithms: RoiEdit3D (panel B), Trackmate (panel C). Green circles represent true positives, while red circles represent false positives. The white bar corresponds to 10 μm.
Fig 4.
ROI with increasing variance of multiplicative noise and resulting accuracy.
Panel A shows different images of the same ROI from data set 2 with an increased variance of multiplicative noise (σ2) from left to right. The length of the white bar represents 10 μm. Panel B shows the resulting plot of the accuracy as a function of noise variance: it is initially relatively stable and then decreases for higher values of noise from about 0.70 to about 0.60.
Table 2.
Comparison of automatic detection performance for different algorithms.
The table shows the number of correctly detected neurons (TP), undetected ones (FN), and false detections (FP), their rates, and the accuracy of the output of See Elegans and two other tracking algorithms: Trackmate and RoiEdit3D. The parameters of See Elegans and TrackMate include threshold and spot size settings, while ROIedit3D was set to the confocal parameter set. See Elegans detects more neurons and has fewer false negatives and positives, outperforming other algorithms.
Table 3.
Performance of the algorithm in the presence of noise.
Percentage of true positives, false negatives, and false positives with increasing variance of multiplicative noise (σ2) applied to a volume before the automatic processing through See Elegans. The increasing level of noise increases the number of false negatives and decreases the number of true positives, thus affecting the accuracy.
Table 4.
Computational time with respect to volume size.
The table shows the computational time required to spot neurons for increasing sizes of volumes. For example, for a 165 x 440 x 12 voxel volume recorded for 10 minutes at 3 volumes per second, the run-time of the detection process is about 7 minutes.
Fig 5.
Tracking of ROIs falling under detection threshold displayed in temporal order.
The top row shows raw data, while the following rows report the tracking results for See Elegans, TrackMate, and ROIedit3D. At t1, neuron 30 is visible in the ROI center, and See Elegans tracks it up to t3, while TrackMate loses it at t2 and ROIEdit3D at t1. The black bar represents 10 μm.
Fig 6.
Resulting traces of neuron 30, falling below the detection threshold.
Panel A shows the traces for ground truth (GT), See Elegans (SE), TrackMate (TM), and ROIedit3D (RE3D). See Elegans captures the dynamics of GT, but TrackMate and RoiEdit3D present gaps and/or artifacts. TrackMate assigns two different IDs to the same neuron, hence the different colors. Panel B reports the absolute displacement of neuron 30 from t1 onwards. Panel C reports the difference between the distance of neuron 30 from 20 of its closest neighbors at t1 and at subsequent times. The plot reveals some neurons moving closer and some moving away with an excursion up to 2 μm (e.g., rows 6 and 16).
Fig 7.
Result of the identification process on data set 2.
Panel A reports the detected neuronal positions rotated to align with a coordinate system in which the x, y, and z axes correspond to the anterior, dorsal and left directions respectively. Panel B shows the spatial arrangement of neurons according to the model. The axes are normalized to the average distance of the 6 initially identified neurons, and the origin is located at the mean point of these neurons. Panels C, D, and E report the trace of the two anti-correlated sets of the neurons, promoting backward (red curves) and forward movement (green curves) respectively.
Table 5.
True positive rates of the automatic identification.
The table shows the true positive rates reached by the algorithm in assigning the neuronal identities to data sets 1,2, and 4–6. The low values of data sets 6 are associated with low signals of the targeted neurons and/or deformation in their typical arrangement.