IDA-MIL: Classification of Glomerular with Spike-like Projections via Multiple Instance Learning with Instance-level Data Augmentation

•Present a novel classification framework by integrating the MIL method and the GAN method for glomerular with spike-like projections.•Propose a new synthetic model InsGAN, which synthesize new positive instances dataset instead of the whole glomerular images.•We give a rigorously analytical study and visualization of inference basis on the proposed framework.•Give the correlation between MN disease stages and high-probability instances proportion based on the historical case.