A Faster Pipeline for Cell Fluorescence Measurements

Neuroscience, Computation  · 

Introduction

This app is an extension to the existing MATLAB tool Measure Cell Fluorescence, an attempt to improve measurement efficiency by optimising manual labelling process wherever possible. However, cell fluorescence measurements rely on the accurate delineation of cellular objects, which is itself a time-consuming process, irreplaceable by a simple computer algorithm. This significantly limited the optimisable components in the procedure.

This work is an open source project in attempt to speed up the ROI delineation by implementing a pre-trained deep learning model. The benefits of deep learning techniques were discussed in a previous project. The deep learning module assists the labelling process while the user may intervene and manually adjust the predicted ROIs. This tool can also be used to generate pixel-level annotations in the form of binary masks, instance masks and MATLAB ROI class objects.

Currently, I am working to deliver the most suitable DCNN model for immunofluorescent-labelled images in the mouse model of Alzheimer’s disease. The pre-release version implements a U-Net + ResNet-34 model architecture. The model was built and trained on Colaboratory with open-source datasets. If you fancy a go at it with your own dataset, you can download the Jupyter notebook here.

Environment Set-up

Measure Corrected Total Cell Fluorescence (CTCF)

  1. Load an image from menu: File > Open… > Image.
  2. Load model, allow up to 20 seconds.
  3. Draw bounding boxes on the image axes to predict cellular ROIs. Note: any feature that touches the borders is ignored.
  4. (Optional) Manually draw additional cellular ROIs, or adjust existing ROIs.
  5. Click the ‘Label’ toggle to switch to begin selecting background regions.
  6. To calculate CTCF, navigate to menu bar: Measure > CTCF
  7. Results are displayed in a table.

Tips: Right click on an ROI object to delete it. Use the label visibility toggle to view the order of ROI objects. If you are not familiar with deep learning, read this before adjusting the threshold knob.

Video Demonstration


Download

The source code will be published in an official release. Version 0.1.1 is now available for download. Version history can be found here. A standalone version which depends on the non-royalty MATLAB runtime will be made available in coming weeks for those who does not have access to a MATLAB license.

Donwload MATLAB App with DCNN Model (262.9MB)

Cite As

Zeyi Yang (2021). Semi-automated Cell Fluorescence Measurement (https://github.com/where-is-brett/cell-fluorescence-ml/releases/tag/v0.1.1), GitHub. Retrieved August 26, 2021.