Tommasino Cristian ; Russo Cristiano ; Rinaldi Antonio Maria

Abstract

In pathology, accurate and efficient analysis of Hematoxylin and Eosin (H\&E)slides is crucial for timely and effective cancer diagnosis. Although many deeplearning solutions for nuclei instance segmentation and classification exist inthe literature, they often entail high computational costs and resourcerequirements, thus limiting their practical usage in medical applications. Toaddress this issue, we introduce a novel convolutional neural network, NuLite,a U-Net-like architecture designed explicitly on Fast-ViT, a state-of-the-art(SOTA) lightweight CNN. We obtained three versions of our model, NuLite-S,NuLite-M, and NuLite-H, trained on the PanNuke dataset. The experimentalresults prove that our models equal CellViT (SOTA) in terms of panoptic qualityand detection. However, our lightest model, NuLite-S, is 40 times smaller interms of parameters and about 8 times smaller in terms of GFlops, while ourheaviest model is 17 times smaller in terms of parameters and about 7 timessmaller in terms of GFlops. Moreover, our model is up to about 8 times fasterthan CellViT. Lastly, to prove the effectiveness of our solution, we provide arobust comparison of external datasets, namely CoNseP, MoNuSeg, and GlySAC. Ourmodel is publicly available at https://github.com/CosmoIknosLab/NuLite