Learn to segment single cells with deep distance estimator and deep cell detector

Abstract

Single cell segmentation is a critical and challenging step in cell imaging analysis. Traditional processing methods require time and labor to manually fine-tune parameters and lack parameter transferability between different situations. Recently, deep convolutional neural networks (CNN) treat segmentation as a pixel-wise classification problem and have become a general and efficient method for image segmentation. However, cell imaging data often possesses characteristics that adversely affect segmentation accuracy: absence of established training datasets, few pixels on cell boundaries, and ubiquitous blurry features. We developed a strategy that combines strengths of CNN and traditional watershed algorithm. First, we trained a CNN to learn Euclidean distance transform (EDT) of the mask corresponding to the input images (deep distance estimator). Next, we trained a faster R-CNN (Region with CNN) to detect individual cells in the EDT image (deep cell detector). Then, the watershed algorithm performed the final segmentation using the outputs of previous two steps. Tests on a library of fluorescence, phase contrast and differential interference contrast (DIC) images showed that both the combined method and various forms of the pixel-wise classification algorithm achieved similar pixel-wise accuracy. However, the combined method achieved significantly higher cell count accuracy than the pixel-wise classification algorithm did, with the latter performing poorly when separating connected cells, especially those connected by blurry boundaries. This difference is most obvious when applied to noisy images of densely packed cells. Furthermore, both deep distance estimator and deep cell detector converge fast and are easy to train.3

DeepCell Kiosk: Scaling deep learning-enabled cellular image analysis with Kubernetes

“…By changing this rate, we simulated different upload speeds. Our image-analysis routine consisted of performing single-cell segmentation using a deep watershed approach 22,23 , which requires both deep learning and conventional processing steps and reveals the relative impact of different computational operations on inference speed and cost. While we benchmarked scalability using the deep watershed approach, we note that this software can (and has) been adapted to deploy a variety of deep learning methods, including RetinaNet 33 and Mask-RCNN 34 , on biological imaging data.…”

“…Resource allocation. Complete computer-vision solutions for cellular image analysis typically require a hybrid of conventional and deep learning methods to achieve a production-ready solution 6,[21][22][23] . We have chosen to separate the conventional and deep learning operations so that they run on different nodes, which allows us to use hardware acceleration for deep learning while ensuring conventional operations are only run on less expensive hardware.…”

Accurate and Versatile 3D Segmentation of Plant Tissues at Cellular Resolution

“…We present results of PlantSeg applications in diverse developmental contexts. PlantSeg is free and open-source, with both a command line and a user-friendly graphical interface.Recently an approach combining the output of two neural networks and watershed to detect individual cells showed promising results on segmentation of cells in 2D [19]. Although this method can in principle be generalized to 3D images, the necessity to train two separate networks poses additional difficulty for non-experts.While deep learning-based methods define the state-of-the-art for all image segmentation problems, only a handful of software packages strives to make them accessible to non-expert users in biology (reviewed in [20]).…”

“…Notably, the U-Net segmentation plugin for ImageJ [21] conveniently exposes U-Net predictions and computes the final segmentation from simple thresholding of the probability maps. CDeep3M [22] and DeepCell [23] enable, via the command-line, the thresholding of the probability maps given by the network, and DeepCell allows instance segmentation as described in [19]. More advanced post-processing methods are provided by the ilastik Multicut workflow [24], however, these are not integrated with CNN-based prediction.Here, we present PlantSeg, a deep learning-based pipeline for volumetric instance segmentation of dense plant tissues at single-cell resolution.…”

“…Recently an approach combining the output of two neural networks and watershed to detect individual cells showed promising results on segmentation of cells in 2D [19]. Although this method can in principle be generalized to 3D images, the necessity to train two separate networks poses additional difficulty for non-experts.…”