PhenoTrack3D

PhenoTrack3D reconstructs and tracks 3D maize plant architecture and individual organ development over time to enable high-throughput temporal phenotyping for genotype-by-environment (GxE) analyses.

Key Features:

• 3D reconstruction and temporal tracking: Extracts a 3D+time (3D+t) reconstruction from time-series datasets and tracks each organ's development over the growth cycle.
• Segmentation (Phenomenal): Leverages existing segmentation methods such as Phenomenal to segment organs in three dimensions as input for tracking.
• Deep-learning stem detection: Uses a deep-learning-based stem detection method to identify the separation point between ligulated and growing leaves with high spatial precision.
• Multiple sequence alignment for ligulated leaves: Implements a multiple sequence alignment algorithm to temporally track, rank, and maintain topology of ligulated leaves.
• Distance-based back-tracking for growing leaves: Applies a distance-based back-tracking approach to follow the development of growing leaves through time.

Scientific Applications:

• Validation on PhenoArch dataset: Validated on 60 maize hybrids imaged daily from emergence to maturity on the PhenoArch platform (~250,000 images), achieving stem tip detection RMSE < 2.1 cm and correctly assigning 97.7% of ligulated leaves and 85.3% of growing leaves across 30 plants over 43 dates.
• Trait extraction and GxE analyses: Enables extraction of organ-level development and architecture traits that show strong correlation with manual observations in random subsets of 10–355 plants, supporting automated characterization for large-scale GxE studies.

Methodology:

Uses Phenomenal 3D segmentation, a deep-learning-based stem detection model, a novel multiple sequence alignment algorithm for ligulated leaves, and a distance-based back-tracking algorithm for growing leaves.