Fig 1.
Workflow for image acquisition, data integration, and annotated dataset generation.
The schematic illustrates the complete pipeline for constructing a labeled dataset for C. elegans behavioral analysis. (Top-left) Original worm images are acquired via microscopy. (Bottom-left) Existing public datasets, as symbolized by the database icon. (Top-right) Representative worm images are displayed. (Bottom-right) Annotation tools are used to label images, generating structured label files (e.g., YOLO-format TXT or JSON annotations). Arrows indicate the sequential flow of data processing and integration.
Fig 2.
Framework for real-time detection and tracking of multiple worms.
The framework consists of two main modules: an object detection network (top, blue) and a tracking and behavior analysis pipeline (bottom).
Fig 3.
The detection module processes input images through a multi-scale backbone and neck for feature extraction and fusion, generating detection results via the prediction head.
Fig 4.
The backbone network of the worm detection module.
Modifying the YOLOv8 detection architecture by integrating the Convolutional Block Attention Module (CBAM).
Fig 5.
Convolutional block attention module.
Fig 6.
The tracking module associates detected targets across frames using Kalman filtering and the Hungarian algorithm, with additional logic to manage trajectory initiation and inactive states. Behavioral analysis is performed based on the resulting trajectories.
Fig 7.
Worm tracking process diagram.
Table 1.
Comparison of different algorithms for worm tracking. All methods were executed on identical hardware (as detailed in Section 2.2) and evaluated on the same dataset using an image size of 1024 × 1024 pixels. The reported results are based on five independent replicates.
Table 2.
Ablation experiment results of the Modified-YOLO model.
Fig 8.
Training result graph of proposed framework.
Fig 9.
Worm detection under different backgrounds.
Fig 10.
Multiple worms tracking trails.
Fig 11.
Movement trajectories, speeds, and movement states of C. elegans.
Fig 12.
A–C shows representative examples corresponding to three bending morphology categories. While these examples illustrate the most commonly observed morphologies, they do not encompass the full spectrum of possible body shapes.
Fig 13.
Roll event detection.
Fig 14.
Omega turn detection of Worms.
Fig 15.
Analysis of multiple advanced locomotor behaviors.
Fig 16.
Heatmaps for combined analysis of worm motion parameters.
Fig 17.
Three-dimensional thermograms for combinatorial analysis of worm motion parameters.
(a)–(c) 3D density distributions of different behavior categories (e.g., reversal turns, rolling, undulation), with color scale indicating high-to-low event frequency across X-position, Y-position, and time. (d) Absence of events in cases of worm inactivity or abnormal states.