Robust Nucleus Detection With Partially Labeled Exemplars

Abstract

Quantitative analysis of cell nuclei in microscopic images is an essential yet challenging source of biological and pathological information. The major challenge is accurate detection and segmentation of densely packed nuclei in images acquired under a variety of conditions. Mask R-CNN-based methods have achieved state-of-the-art nucleus segmentation. However, the current pipeline requires fully annotated training images, which are time consuming to create and sometimes noisy. Importantly, nuclei often appear similar within the same image. This similarity could be utilized to segment nuclei with only partially labeled training examples. We propose a simple yet effective region-proposal module for the current Mask R-CNN pipeline to perform few-exemplar learning. To capture the similarities between unlabeled regions and labeled nuclei, we apply decomposed self-attention to learned features. On the self-attention map, we observe strong activation at the centers and edges of all nuclei, including unlabeled nuclei. On this basis, our region-proposal module propagates partial annotations to the whole image and proposes effective bounding boxes for the bounding box-regression and binary mask-generation modules. Our method effectively learns from unlabeled regions thereby improving detection performance. We test our method with various nuclear images. When trained with only 1/4 of the nuclei annotated, our approach retains a detection accuracy comparable to that from training with fully annotated data. Moreover, our method can serve as a bootstrapping step to create full annotations of datasets, iteratively generating and correcting annotations until a predetermined coverage and accuracy are reached. The source code is available at https://github.com/feng-lab/nuclei. INDEX TERMS Nucleus segmentation, deep learning, convolutional neural networks, few-exemplar learning, semisupervised learning, computer-assisted annotating.

Topographic connectivity and cellular profiling reveal detailed input pathways and functionally distinct cell types in the subthalamic nucleus

“…A nucleus detection pipeline, as previously described (Feng et al, 2019), was trained on 30 images containing DAPI-stained cell nuclei ($180 per image) that were manually labeled by human experts. To detect mRNA puncta, we used a splitting strategy with a marker-controlled watershed approach and a variational Bayesian Gaussian mixture model (Feng et al, 2012).…”

A survey on recent trends in deep learning for nucleus segmentation from histopathology images

“…The measured values of ZSI = 0.933 ± 0.14, Precision = 0.946 ± 0.06 and Recall = 0.984 ± 0.00 Limitations: Since the method is based on a U-shaped network, a skip connection arises. The model would not have performed better due to the skip connection Year: 2019 Feng et al ( 2019 ) propose a region-proposal module to perform exemplar learning Features: Backbone: ResNeXt-101–64 × 4d Loss: Classification Loss was Cross-entropy loss and Regression Loss is Smooth L1 The proposed model used a self-attention mechanism to capture the similarity between nuclei and to strongly handle moderately labelled training images, The framework wherein Region Proposal Network (RPN) to produce a large number of bounding box candidates to densely cover the image. Furthermore, an object score for an individual candidate is calculated Comparison: DIST and CNN Model with Three Convolutional Layers (CNN-3) Datasets: Cell images of seven distinct organs (breast, kidney, Liver, Prostate, Bladder, Colon, and Stomach) were extracted from Nucleus Dataset (Cicconet et al 2017 ), Haematoxylin and Eosin (H&E)-stained Histopathology Dataset (Naylor et al 2018 ) Parameters: Aggregated Jaccard Index (AJI) Inference: The proposed model was better in terms of performance as compared to other models as it delivered a better solution when extreme segmentation accuracy was desirable by using it as a computer-assisted annotation tool to produce high-quality fully annotated training datasets.…”

Topographic connectivity and cellular profiling reveal detailed input pathways and functionally distinct cell types in the subthalamic nucleus

“…We used probes for Aqp4 (catalog no. A nucleus detection pipeline, as previously described (Feng et al, 2019), was trained on 30 images containing DAPI-stained cell nuclei (~180 per image) that were manually labeled by human experts. To detect mRNA puncta, we used a splitting strategy with a marker-controlled watershed approach and a variational Bayesian Gaussian mixture model (Feng et al, 2012).…”