Top Highlights

• We employed a multiple linear regression model, considering the method’s mAP as the dependent variable, and the number of training images, cell elongation, and cell size as independent variables (Methods). Cell convexity was excluded due to its minimal variation across tissues (Supplementary Figure 4). For most methods and segmentation tasks, there...
• We then explored a strategy of segmenting images using the same modality as the training data, followed by morphologically dilating cell nuclei segmentation to approximate whole cell segmentation, or eroding whole cell segmentation for cell nuclei segmentation (Methods

Domain