Abstract
AbstractMotivationIn the past decade, deep learning algorithms have surpassed the performance of many conventional image segmentation pipelines. Powerful models are now available for segmenting cells and nuclei in diverse 2D image types, but segmentation in 3D cell systems remains challenging due to the high cell density, the heterogenous resolution and contrast across the image volume, and the difficulty in generating reliable and sufficient ground truth data for model training. Reasoning that most image processing applications rely on nuclear segmentation but do not necessarily require an accurate delineation of their shapes, we implemented PAC-MAP, a 3D U-net based method that predicts the position of nuclei centroids and their proximity to other nuclei.ResultsWe show that our model outperforms existing methods, predominantly by boosting recall, especially in conditions of high cell density. When trained from scratch PAC-MAP attained an average F1 score of 0.793 in dense spheroids. When pretraining using weakly supervised bulk data input and finetuning with few expert annotations the average F1 score could be significantly improved up to 0.817. We demonstrate the utility of our method for quantifying the cell content of spheroids and mapping the degree of glioblastoma multiforme infiltration in cerebral organoids.Availability and implementationThe code is available on GitHub, athttps://github.com/DeVosLab/PAC-MAP.ContactWinnok H. De Vos (winnok.devos@uantwerpen.be)
Publisher
Cold Spring Harbor Laboratory