AbstractBackgroundResearchers interested in the seed size of rapeseed need to quantify the cell size and number of cells in the seed coat, embryo and silique wall. Scanning electron microscope-based methods have been demonstrated to be feasible but laborious and costly. After image preparation, the cell parameters are generally evaluated manually, which is time consuming and a major bottleneck for large-scale analysis. Recently, two machine learning-based algorithms, Trainable Weka Segmentation (TWS) and Cellpose, were released to overcome this long-standing problem. Moreover, the MorphoLibJ and LabelsToROIs plugins in Fiji provide user-friendly tools to deal with cell segmentation files. We attempted to verify the practicability and efficiency of these advanced tools for various types of cells in rapeseed.ResultsWe simplified the current image preparation procedure by skipping the fixation step and demonstrated the feasibility of the simplified procedure. We developed three methods to automatically process multicellular images of various tissues in rapeseed. The TWS–Fiji (TF) method combines cell detection with TWS and cell measurement with Fiji, enabling the accurate quantification of seed coat cells. The Cellpose–Fiji (CF) method, based on cell segmentation with Cellpose and quantification with Fiji, achieves good performance but exhibits systematic error. By removing border labels with MorphoLibJ and detecting regions of interest (ROIs) with LabelsToROIs, the Cellpose–MorphoLibJ–LabelsToROIs (CML) method achieves human-level performance on bright-field images of seed coat cells. Intriguingly, the CML method needs very little manual calibration, a property that makes it suitable for massive-scale image processing. Through a large-scale quantitative evaluation of seed coat cells, we demonstrated the robustness and high efficiency of the CML method at both the single-cell level and the sample level. Furthermore, we extended the application of the CML method to developing seed coat, embryo and silique wall cells and acquired highly precise and reliable results, indicating the versatility of this method for use in multiple scenarios.ConclusionsThe CML method is highly accurate and free of the need for manual correction. Hence, it can be applied for the low-cost, high-throughput quantification of diverse cell types in rapeseed with high efficiency. We envision that this method will facilitate the functional genomics and microphenomics studies of rapeseed and other crops.