Fig 1.
Dynamism in video microscopy of T. cruzi parasites.
Three motions are illustrated: (A) collateral motion observed during the locomotion of T. cruzi parasites between cells; (B) the fluctuating motion of cells or other artifacts present in the blood; and (C) PTZ motion due to microscope or camera focus adjustments.
Fig 2.
Overview of trajectory analysis.
(A) The trajectories of parasites and cells are split to form training and test datasets. (B) The trajectories are partitioned into segments for a better analysis of their behavior. (C) Features are extracted from trajectory segments. (D) Approach to discriminate between elements. Supervised learning classification models learn the training dataset’s motion patterns and categorize the test dataset’s trajectory segments. (E) Approach to identify cell motion patterns. Unsupervised learning models identify hidden patterns in training and test trajectory segments and group similar segments into a cluster. (F) We evaluated the performance of the experiments according to the classification accuracy (distinguishing between parasites and cells) and the clustering quality (distinguishing between fluctuating and PTZ motion).
Fig 3.
Trajectory segments of the T. cruzi parasite obtained from the sequential and random splitting strategies.
(A) Sequential split: segments are generated sequentially along the complete trajectory at regular intervals of k points. Different colors in the complete trajectory represent the segments. The parasite region is highlighted for some trajectory segments. (B) Random split: segments are obtained by randomly choosing a starting point on the trajectory and the k − 1 following points.
Fig 4.
Feature analysis from training trajectory segments.
(A) Pairwise correlation of features. The values correspond to the correlation coefficient: -1.00 indicates a perfect negative correlation, +1.00 indicates a perfect positive correlation, and 0 indicates no correlation between the features. (B) Distribution and probability density of trajectory segments by feature. Blood cell and parasite segments are represented by pink and green rhombuses, respectively. White rhombuses indicate the median of the distribution. The width of each curve corresponds to the approximate frequency of trajectory segments in the region. (C) Mean squared displacements (MSDs) versus the number of video frames. The curves summarize the computed MSD at intervals of 1 to 10 video frames (∀j ∈ [1, 10] in MSD(tj)) for all trajectory segments.
Table 1.
Accuracy results for test trajectory segments for different tree-based classifiers.
Table 2.
XGBoost classifier performance metrics.
Fig 5.
Qualitative analysis of the XGBoost classifier results in the feature collection experiment.
(A) Trajectory segments by predicted class. The lighter tones of each segment indicate the beginning of the trajectory, and the darker tones indicate the end. The moving element region is highlighted for some trajectory segments. The arrow indicates the precise location of the element. (B) Comparison of the distance traveled by the elements.
Table 3.
Spectral clustering results for the dataset of cell trajectory segments.
Fig 6.
Clustering analysis of blood cell trajectory segments.
Clusters obtained in the experiments: (A) the feature collection, in which the moving cell region is highlighted for some trajectory segments; (B) TSD + λ-MSD; (C) TSD + mean speed; and (D) λ-MSD + mean speed. (E) Behavior of silhouette coefficient and percentage of bi-labeled trajectories across different numbers of clusters for the studied feature combinations. Notably, the experiments reached the best results with two clusters, as we observed the highest silhouette coefficient value in conjunction with the lowest percentage of bi-labeled trajectories.
Fig 7.
Analysis of bi-labeled trajectories in the feature collection experiment.
(A) Complete cell trajectories with labeled steps according to segment clustering. The trajectories of cells c1 to c6 are analyzed below. (B) Influence of dynamic events on the occurrence of bi-labeled trajectories. The graph’s axes correspond to the trajectory segments, and the lines of specific colors represent the cells. The color of the rhombus indicates the clustering label. Dynamic events, which can focus on adjustments or dynamic fluids, are signaled at the end of each axis. We also highlight the frame number in which the event started and ended.