A deep learning-integrated micro-CT image analysis pipeline for quantifying rice lodging resistance-related traits

Lodging is a common problemin rice,reducing its yield andmechanical harvesting efficiency.Rice architecture is a key aspect of its domestication and a major factor that limits its high productivity.The ideal rice culm structure,includingmajor_axis_culm,minor axis_culm,andwall thickness_culm,is critical for improving lodging resistance.However,the traditionalmethod ofmeasuring rice culms is destructive,time consuming,and labor intensive.In this study,we used a high-throughput micro-CT-RGB imaging system and deep learning(SegNet)todevelopa high-throughputmicro-CTimageanalysis pipelinethatcanextract 24 riceculmmorphological traits and lodging resistance-related traits.When manual and automatic measurements were compared at themature stage,the mean absolute percentage errors for major_axis_culm,minor_axis_culm,andwall_thickness_culmin 104 indica rice accessionswere 6.03%,5.60%,and 9.85%,respectively,and the R^(2) valueswere 0.799,0.818,and 0.623.We also builtmodels of bending stress using culmtraits at the mature and tillering stages,and the R^(2) values were 0.722 and 0.544,respectively.The modeling results indicated that this method can quantify lodging resistance nondestructively,even at an early growth stage.In addition,we also evaluated the relationships of bending stress toshoot dryweight,culm density,and drought-related traits and found that plants with greater resistance to bending stress had slightly higher biomass,culm density,and culm area but poorer drought resistance.In conclusion,we developed a deep learning-integrated micro-CT image analysis pipeline to accurately quantify the phenotypic traits of rice culms in4.6 min per plant;this pipeline will assist in future high-throughput screening of large rice populations for lodging resistance.