Icy, an open source image analysis platform, also provides a plug-in for viewing and editing tracks (de Chaumont et al., 2012). Performance evaluation, also referred as performance analysis, in image analysis compares the results obtained from an automated procedure against the manually established ‘ground truth’. Herein, a ground truth track represents
the ‘true’ positions of a cell as a sequence of bounding boxes. We used the Video Performance Evaluation Resource (ViPER) software (Doermann and Mihalcik, 2000) to manually draw bounding boxes around cells in each video frame and index the sequences of bounding boxes corresponding JQ1 in vitro to each individual cell to designate tracks. Performance evaluation metrics were employed to quantitatively and comprehensively assess the detection and tracking performance of TIAM and the third-party tools. We used the Sequence Frame Selleckchem RO4929097 Detection Accuracy (SFDA) and Average Tracking Accuracy (ATA) metrics (Kasturi et al.,
2009) as these can be computed in a fully automated fashion and thus allow for reproducible quantification of the success of detection and tracking of objects. Further, they do not suffer from the risk of human error or bias. These metrics have been adopted as standardized metrics by the Video Analysis and Content Extraction (VACE) program (http://marathon.csee.usf.edu/vace-links.html) and the Classification of Events, Activities, and Relationships
(CLEAR) consortium (www.clear-evaluation.org); which are two large-scale and community-wide efforts concerned with video tracking and interaction analysis. The metrics are based on Jaccard Similarity (Fig. S5 for intuitive illustration and and Supplementary methods for mathematical description). In order to compute SFDA and ATA, a one-to-one correspondence between ground truth Etomidate and result must be established. To establish this mapping we employed the Hungarian algorithm (Munkres, 1957) with metrics based on Jaccard Similarity used to construct the similarity matrix (see Supplementary methods for details). We have consolidated the software routines to carry out performance analysis in a separate MATLAB-based suite that we call PACT (Performance Analysis of Cell Tracking). The PACT code, its user guide and relevant ground truth datasets are available at https://github.com/willieneis/TIAM/tree/master/PACT/. The user guide also includes specific instructions on using ViPER for ground truth annotation. Performance of feature extraction was also evaluated against ground truth. Outlines drawn manually or by semi-automated procedures in ImageJ (Schneider et al., 2012) were listed as ROIs and used as ground truth (see Supplementary methods for details). A one-to-one correspondence between individual cells in ground truth and TIAM result was obtained using the Hungarian algorithm (Munkres, 1957).