Automatic 3D cell segmentation of fruit parenchyma tissue from X-ray micro CT images using deep learning

Area of Science:

• Plant biology
• Biophysics
• Medical imaging

Background:

• High-quality 3D plant tissue morphology is crucial for understanding plant physiology.
• X-ray micro-CT provides 3D microstructural data but struggles with segmenting individual cells due to low density contrast.
• Accurate cell segmentation is needed to analyze cell morphology and spatial organization in plant tissues.

Purpose of the Study:

• To develop and evaluate deep learning models for segmenting individual plant cells in 3D using X-ray micro-CT data.
• To compare the performance of deep learning models against existing methods for plant tissue analysis.
• To investigate the influence of tissue characteristics on segmentation accuracy.

Main Methods:

• Trained and tested deep learning-based models on X-ray micro-CT images of apple and pear parenchyma tissues.
• Utilized parenchyma tissue samples with varying cell and porosity characteristics.
• Evaluated segmentation performance using the Aggregated Jaccard Index (AJI).

Main Results:

• The best deep learning model achieved AJIs of 0.86 for apple and 0.73 for pear, outperforming the benchmark method (0.73 and 0.67).
• The model successfully identified other structures like vascular bundles and stone cell clusters, revealing their impact on cell organization.
• Segmentation accuracy was higher for apple tissue due to its higher porosity and lower specific surface area compared to pear tissue.

Conclusions:

• The proposed deep learning method enables automated 3D cell morphology quantification from micro-CT, replacing manual or less accurate approaches.
• Challenges in segmentation arise with low fruit tissue porosity, low pore network connectivity, or high pore space surface area.
• For difficult samples, advanced contrast-enhancing scan protocols may be necessary for accurate cell outline detection.