This is a Python 3 / PyQt5 project dedicated to the analysis of the images generated by the AMi microscope. The AMi microscope is a low-cost automated microscope.
For more details on this device, please look at: "AMi: a GUI-based, open-source system for imaging samples in multi-well plates" [1]
This program should work with images taken by other machines given the folder tree is as expected.
I created this application to help the members of our laboratory in their crystallization work. The software can read tiff, jpeg or png files. Images are accessed via clicking on the corresponding well button or by navigating with keyboard shortcuts. Images can be zoomed for thorough inspection and a timeline of the corresponding well is displayed if several pictures of the same well taken at different times are available.
Images are scored using the Drop Score section. Notes can be taken and are saved to file.
A report of the current well can be saved as a pdf.
The results can be displayed on a grid and statistics are also calculated.
The tree must be organized like:
.
└── images
└── sample directory (used as Project Code)
└── plate directory (used as Plate name)
└── YYYYMMDD_HHMMSS
├── stacked images (generated from directory rawimages)
└── rawimages
└── files at different focus height
├── YYYYMMDD_HHMMSS
├── ...
└── prep_date.txtAMi controlling software needs a folder named "images". The date directory must have the format YYYYMMDD_HHMMSS (eg. 20191211_151023) The file "prep_date.txt" contains the preparation date of the plate. It contains a single line with a date with format YYYYMMDD and is created within the program. It is used for logging purpose only and is not mandatory.
2019/12/12: Automated annotation using TensorFlow (https://www.tensorflow.org/) and MARCO (https://marco.ccr.buffalo.edu/) has been added. For now, this requires TensorFlow version previous to v2.
Important Note: a threshold is applied to accept or not autoMARCO most probable prediction. If the prediction probability is below this threshold (currently 0.6 and can be modified in preferences.py), the drop classification is set to "Unknown".
2019/12/13: Added visualisation of autoMARCO results
2019/12/17: Added a tool to automatically crop images using opencv (autocrop.py). This enhances autoMARCO results. Added a tool to check circle detection on a given image (Check_Circle_detection.py).
2020/01/10: Parameters controlling circle detection are set in preferences.py The color of the field indicating the actual classification changes dynamically. New release.
2020/01/16: Changed startup routine to include python virtual environment initialisation. Release 1.1.6.
2020/01/23: Fixed autoMARCO on Raspbian Buster. Fixed bug leading to program crash when stopping autoMARCO and filtering results. Release 1.1.8.
2020/01/29: Updated tensorflow to 1.15.2 in requirements.txt
2020/02/03 Images in Timeline can be clicked to load corresponding image in main viewing frame for closer inspection. autoMARCO windows can all be displayed at the same time (previously only a single window could be loaded)
2020/03/03 Merging Z-stacks can be done within AMI_Image_Analysis or command line using the script Merge_Zstack.py. It is done in python using openCV and is much faster than using Hugin.
2020/05/20 Improved navigation with shortcuts: Now only visible wells through "Filter" option are selected. Added Screens Tables to quickly check crystallization conditions. Crystallization Tables are read from csv files stored in the folder "Screen_Database".
The software was tested on:
- linux CentOS 7 (PyQt5 (v5.9.2 and v5.12.1) and python3 (v3.6.8 and v3.7.3))
- Raspbian 10 (on a Pi 3B+, not recommended)
- macOS HighSierra and Catalina
Hopefully, it should work on other linux distributions, MacOS and Windows. You will need a screen with minimum resolution of 1920*1080.
It is recommended to create a virtual environment for python (see INSTALL.txt)
Download the latest released version and gunzip it or clone this repository with git clone https://github.com/LP-CDF/AMi_Image_Analysis AMi_Image_Analysis. Then follow the instructions in INSTALL.txt
Dependencies:
- Qt5
- PyQt5 (v5.9.2 or higher)
- Python Imaging Library (Pillow, v6.2.1 or higher)
- PyFPDF (v1.7.2 or higher)
Optional dependencies:
- TensorFlow (v1.14 or 1.15, not v2)
- MARCO tensorflow model (https://storage.googleapis.com/marco-168219-model/savedmodel.zip) One version of this model is included in saved_model/ Methodology details are published in [2]
- openCV for the automatic cropping tool (tested with opencv 4.0.1)
- NumPy
Note: a pip requirement file is available for easy install of python dependencies.
Known issues:
- On Raspbian AutoMARCO does not work with regular PyPi tensorflow package (see INSTALL.txt).
- If your CPU does not support AVX instruction sets (CPU before SandyBridge), you will need to find a tensorflow with the correct building options (have a look here)
If you wish to cite this work, you can use the following reference: To be added
This project would not have been possible without the previous work of Dakota Handzlik published in [3]. Corresponding project can be found at: https://github.com/dakota0064/Fluorescent_Robotic_Imager
All the Python, Qt communities for sharing experience.
[1] "AMi: a GUI-based, open-source system for imaging samples in multi-well plates" Andrew Bohm, Acta Crystallogr F Struct Biol Commun. 2019 Aug 1; 75(Pt 8): 531–536. doi: 10.1107/S2053230X19009853
[2] "Classification of crystallization outcomes using deep convolutional neural networks." Bruno AE et al. PLoS One. 2018 Jun 20;13(6):e0198883. doi: 10.1371/journal.pone.0198883
[3] "Inexpensive robotic system for standard and fluorescent imaging of protein crystals." Handzlik D et al. Acta Crystallogr F Struct Biol Commun. 2019 Nov 1;75(Pt 11):673-686. doi: 10.1107/S2053230X19014730