Table 1 Comparison of LIM Tracker with those of the existing software.

		TrackMate (ImageJ/Fiji)	CellProfiler (Broad Institute)	MMHelper (Univ. of Exeter)	Usiigaci (OIST)	LIM Tracker
Comparison of the function
Tracking function	Link-type tracking	Yes	Yes	Yes	Yes	Yes
	Sequential tracking	No	No	No	No	Yes
	Manual tracking	Limited^a	No	No	No	Yes
Interactive real-time data linkage display		No	No	No	No	Yes
ROI editing function		Yes^b	No	No	No	Yes^c
Trajectory editing function		Limited^d	No	No	No	Yes
Recognition function (non-DL)		Yes	Yes	Yes	No	Yes
DL recognition function	Recognition function	Yes	No	No	Yes	Yes
	Training function (integrated UI including annotation)	No	No	No	Limited^e	Yes
Performance on ISBI Cell Tracking Challenge⁶ “PhC-C2DH-U373” dataset (Fig. 5c)
Recognition accuracy (SEG) ^f		0.68^g	N/A^h	N/A ⁱ	0.89^j	0.93^k
Detection accuracy (DET) ^l		0.93	N/A	N/A	0.97	0.98
Tracking accuracy (TRA) ^m		0.93	N/A	N/A	0.97	0.98

a, Cannot be combined with other tracking functions. b, The region shape cannot be set. c, The region shape can also be freely set. d, It only works for a very small number of targets. e, A Python script file is provided, and executed by command line operations. f, l, m, The evaluation index is the one used in the ISBI Cell Tracking Challenge⁶, and is calculated using a publicly available evaluation program. g, Use the DL recognition function “Stardist (Stardist detector custom model)”²³. Since it has no training function, it creates trained weight files based on command line operations. h, i, It did not have a high-performance recognition function and could not be recognized. j, Use the DL recognition function “Mask R-CNN”²⁵. The included Python script is used to create a trained weight file based on command line operations. k, Use the DL recognition function “Mask R-CNN”²⁵, and the training function (integrated UI including annotation) enables efficient training with simple operations.

Quick links

Search