An enhanced loss function simplifies the deep learning model for characterizing the 3D organoid models

Abstract

Motivation Organization of the organoid models, imaged in 3D with a confocal microscope, is an essential morphometric index to assess responses to stress or therapeutic targets. In fact, differentiating malignant and normal cells is often difficult in monolayer cultures. But in 3D culture, colony organization can provide a clear set of indices for differentiating malignant and normal cells. The limiting factors are delineating each cell in a 3D colony in the presence of perceptual boundaries between adjacent cells and heterogeneity associated with cells being at different cell cycles. Results In a previous paper, we defined a potential field for delineating adjacent nuclei, with perceptual boundaries, in 2D histology images by coupling three deep networks. This concept is now extended to 3D and simplified by an enhanced cost function that replaces three deep networks with one. Validation includes four cell lines with diverse mutations, and a comparative analysis with the UNet models of microscopy indicates an improved performance with the F1-score of 0.83. Availability All software and annotated images are available through GitHub and Bioinformatics online. The software includes the proposed method, UNet for microscopy that was extended to 3D, and report generation for profiling colony organization. Supplementary information Supplementary data are available at Bioinformatics online and Github.

Marker-controlled watershed with deep edge emphasis and optimized H-minima transform for automatic segmentation of densely cultivated 3D cell nuclei

“…Another highlight of this study was the comparison of 3D version of U-Net-Cell [29,35] against {B, M 3DE , −} configuration. The configuration was otherwise the same as U-Net-Cell but did not utilize weight maps during training and produced instance segmentation using H-minima transform-based watershed algorithm.…”

“…Similar results were achieved over independent datasets with average PQ improving from 0.61 to 0.71. In a previous study with 3D nuclei dataset, U-Net-Cell was compared to an approach similar to U-Net-Cell but which used an enhanced loss function during training [35]. The use of the enhanced loss function led only to modest improvements as average AJI score improved from 0.63 to 0.66.…”

“…Additionally, we devised 3D versions of U-Net proposed in [29]. These versions, named as U-Net-Cell and U-Net-Cell* similarly as in [35], were similar to {B, M 3DE , −} and {B, M 3DE , −, * } but with two main differences: instance segmentations were produced using connected component analysis and the loss function used in training of the edge masks was a 3D version of weighted cross entropy loss introduced in [29]. In this loss function, voxels representing boundaries of clumped nuclei were given higher weight.…”

“…This, however, can be problematic, as a misdetection of single boundary pixel may lead to the fusion of touching nuclei. To alleviate this problem, various loss functions which emphasize cell boundaries [29,32,34,35] and the use of edge emphasizing masks alongside watershed algorithm have been proposed [25,34,36]. However, modified loss functions and improved neural network architectures proposed in the literature almost always include tuning of a set of hyperparameters.…”

Application of medical imaging methods and artificial intelligence in tissue engineering and organ-on-a-chip

“…Kong et al use machine learning methods in colorectal and bladder organoid models to predict the efficacy of anti-cancer drugs in patients ( Kong et al, 2020 ). In addition, researchers have improved deep learning methods for characterizing organoid models using augmented loss functions based on previous studies ( Winkelmaier and Parvin, 2021 ). The development of organoids and OOC is unstoppable, and the application of artificial intelligence methods will undoubtedly bring greater vitality and impetus to the development of this field.…”