Differential cell counts using center‑point networks achieves human‑level accuracy and efficiency over segmentation

published in Scientific Report on 19 August 2021

Setup

1. In order to run this notebook, please follwing the fast.ai environment setup instruction. https://docs.fast.ai/#Installing

2. git clone this repository

3. download images (.jpg and .pkl), annotations (.txt) and model weighted (.pth) from Google Drive.

4. Before running the notebook, arrange files as follows:

   \DCNet\(notebook.ipynb)
     \(annotations.txt)
     \(images.pkl)
     \model\(model weight.pth)

Model Architecture

Schematic diagram (generated by Netron)

see dcnet-resnet34-entire-model.onnx.svg

Supplementary information

All the data and models used for this project can be found here:

Full data + models: https://drive.google.com/file/d/1JxSfyzxZlqUtoN_JPUzTzWq7_qPj95zw/view?usp=sharing
Data Only (Cytospin + KDSB): https://drive.google.com/file/d/1Mx11mSoGq-pYkivzayG0hvSedMBsKRBG/view?usp=sharing
Mask-RCNN (ablation model): https://drive.google.com/file/d/1REq4UUfKk3tuKn7Ks42xHY18IjedR7_Y/view?usp=sharing

Cite this article

TY  - JOUR
AU  - Lee, Sarada M. W.
AU  - Shaw, Andrew
AU  - Simpson, Jodie L.
AU  - Uminsky, David
AU  - Garratt, Luke W.
PY  - 2021
DA  - 2021/08/19
TI  - Differential cell counts using center-point networks achieves human-level accuracy and efficiency over segmentation
JO  - Scientific Reports
SP  - 16917
VL  - 11
IS  - 1
AB  - Differential cell counts is a challenging task when applying computer vision algorithms to pathology. Existing approaches to train cell recognition require high availability of multi-class segmentation and/or bounding box annotations and suffer in performance when objects are tightly clustered. We present differential count network (“DCNet”), an annotation efficient modality that utilises keypoint detection to locate in brightfield images the centre points of cells (not nuclei) and their cell class. The single centre point annotation for DCNet lowered burden for experts to generate ground truth data by 77.1% compared to bounding box labeling. Yet centre point annotation still enabled high accuracy when training DCNet on a multi-class algorithm on whole cell features, matching human experts in all 5 object classes in average precision and outperforming humans in consistency. The efficacy and efficiency of the DCNet end-to-end system represents a significant progress toward an open source, fully computationally approach to differential cell count based diagnosis that can be adapted to any pathology need.
SN  - 2045-2322
UR  - https://doi.org/10.1038/s41598-021-96067-3
DO  - 10.1038/s41598-021-96067-3
ID  - Lee2021
ER  - 

Disclaimer

https://www.freeprivacypolicy.com/live/5a5dff6f-592a-4271-8fcc-8dbf729d1171