RootPainter

RootPainter enables deep learning segmentation of biological images to train convolutional neural networks for quantitative analysis such as root length extraction, biopore counting, and root nodule counting.

Key Features:

• Deep learning segmentation: Performs image segmentation using deep neural networks for biological image analysis.
• Convolutional neural networks (CNNs): Uses CNN architectures to learn features for segmentation and measurement tasks.
• Corrective annotations: Supports corrective annotations added during the training phase based on identified weaknesses of the current model.
• Iterative model refinement: Enables iterative refinement of models by incorporating corrective annotations during training.
• Rapid training: Enables rapid training of deep neural networks tailored to specific tasks, with models developed within short annotation times in reported evaluations.
• Evaluation on plant roots: Has been evaluated on chicory (Cichorium intybus L.) roots in soil with quantitative comparison to manual measurements.
• Annotation–accuracy relationship: Demonstrates a correlation between model accuracy and duration of annotation in empirical studies.
• Robustness across image conditions: Shows applicability across datasets with varying target objects, backgrounds, and image qualities.

Scientific Applications:

• Root length extraction: Quantifies root length from segmented root images for plant phenotyping studies.
• Biopore counting: Detects and counts biopores in soil images through segmentation-based analysis.
• Root nodule counting: Identifies and counts root nodules via trained segmentation models.
• Chicory root measurement: Applied to chicory (Cichorium intybus L.) roots in soil, producing models that correlated strongly with manual measurements in five out of six cases after approximately two hours of annotation.

Methodology:

Training uses corrective annotations added during the training phase to iteratively refine convolutional neural network segmentation models.