Phenotyping alfalfa (Medicago sativa L.) root structure architecture via integrating confident machine learning with ResNet-18

Authors: Weihs, Brandon; TANG, ZHOU - Washington State University; TIAN, ZEZHONG - University Of Wisconsin; Heuschele, Deborah - Jo; SIDDIQUE, AFTAB - Fort Valley State University; TERRILL, THOMAS - Fort Valley State University; ZHANG, ZHOU - University Of Wisconsin; YORK, LARRY - Oak Ridge National Laboratory; ZHANG, ZHIWU - Washington State University; Xu, Zhanyou

Submitted to: Plant Phenomics
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/20/2024
Publication Date: 9/11/2024
Citation: Weihs, B.J., Tang, Z., Tian, Z., Heuschele, D.J., Siddique, A., Terrill, T., Zhang, Z., York, L., Zhang, Z., Xu, Z. 2024. Phenotyping alfalfa (Medicago sativa L.) root structure architecture via integrating confident machine learning with ResNet-18. Plant Phenomics. https://doi.org/10.34133/plantphenomics.0251.
DOI: https://doi.org/10.34133/plantphenomics.0251

Interpretive Summary: Roots are the nutrient and acquisition organs for most plants, and improvements of plant roots are important for increasing aboveground traits such as biomass yield or increased nutrition. A modern method for investigating plants roots involves pairing labeled images with artificial intelligence (AI) to detect and predict root properties such as root length, the number of daughter roots, or the entire root system architecture. For this study, root system architectures of alfalfa plants were analyzed using images of root systems input into an AI model. The model results were improved from11-13% by applying two additional methods called reactive machine learning and confident machine learning. These results are important to alfalfa breeders, root scientists, and producers because using AI to detect root traits and architecture will help to remove the human bias involved in breeding selection efforts as well as increase the speed and efficiency of the trait selection process over manual methods (by hand). By using this high-accuracy AI model and image label correction method, the desired traits for root breeding are more easily achieved and in a more time-efficient and non-biased manner.

Technical Abstract: Background – Root system architecture (RSA) is of growing interest in implementing plant improvements with belowground root traits. Modern computing technology applied to images offers new pathways forward to plant trait improvements and selection through RSA analysis. However, image label noise reduces the accuracies of models. This study utilized an AI model capable of classifying RSA directly from images. Images were compared with different model outputs with manual root classifications, then tested confident machine learning (CL) and reactive machine learning (RL) methods to minimize the effects of subjective labeling and improve labeling and prediction accuracies. Results – The CL algorithm modestly improved the Random Forest model’s overall prediction accuracy of the MN data set (1%) while larger gains in accuracy were observed with the ResNet-18 model results. The ResNet-18 cross-population prediction accuracy was improved (~8 to 13%) with CL compared to the original/uncorrected datasets. Training and testing data combinations with the highest accuracies (86%) resulted from the CL and/or RL corrected datasets for predicting taproot RSAs. Similarly, the highest accuracies achieved for the intermediate RSA class resulted from corrected data combinations. The highest overall accuracy (~75%) using the ResNet-18 model involved CL on a pooled dataset containing images from both sample locations. Conclusions – ResNet-18 DNN prediction accuracies of alfalfa RSA images are increased when CL and RL are employed. By increasing the dataset to reduce overparameterization while concurrently finding and correcting image label errors, it is demonstrated here that accuracy increases as much as ~11-13% can be achieved with semi-automated, computer-assisted preprocessing and data cleaning (CL/RL).