This project focuses on the analysis and classification of fluorescence microscopy images of mouse fibroblast cells irradiated with either Fe particles or X-ray, and the characterization of DNA damage indicated by 53BP1 foci presence related to radiation exposure.
We are analyzing the NASA Open Science Data Repository: OSD-366 (DOI: 10.26030/v8w4-rg83) available here:
https://registry.opendata.aws/bps_microscopy/
Thank you, I will contact @evartsb
Thanks for the info. We’re also working on the same OSDR and have just begun manual quantification of the foci.
Is there a specific tool which is user friendly and efficient at the same time to label the foci?
Best,
Anu
Hi @anuiris this might be helpful: Meta’s Segment Anything Model (SAM). https://segment-anything.com/
@evartsb i don’t know how to access the images via the AWS S3 bucket, so if you could attach some sample images from several strains here that would be great.
My company’s software can measure shapes in very unique ways.
Example:
Video [15 min]: Shape Analysis Detects Malaria Without Giemsa Staining
Slides from video:
Thank you very much. Let us try with this.
Hi @anuiris ! I evaluated several tools and the issue I came across was .tif file format compatibility with many labeling tools. I ended up finding a python script that runs in Jupyter Lab that I was able to modify to fit our purposes. If you come across any free tools that can be used to quickly label a large quantity of images, please let me know!
Hi @dave ! I used aws cli to download the files.
steps:
I hope this helps! Let me know if you have any issues and as always, stack overflow can probably explain it 100 ways better than I can 
Hi @anuiris Anu, I have developed different tools to quantify foci. What coding language do you use?
We had the same issue. However, my colleague Dr Josef has now written a script for a tool that allows us to mark the foci quite well. We’re in the process of refining it. There’s a group from Belgium that suggested Focinator but I haven’t tried it. As of now, I have LabelMe which isn’t quite user friendly.
Hi, we were about to send you an email regarding this @svcostes
Dr Gianluca is building the AI model, Dr Josef Borg has now created a tool to help with identification of the foci. They’re using Python.
My current role is in manually labelling the images and LabelMe was suggested which is not quite efficient in my opinion.
Do you have an easier and efficient tool for manually labelling the foci??
Thanks in advance!
Hi Anu
@evartsb has a labeling tool. For spot detection, I recommend to stay away from threshold and top hat. Wavelet works best. I’m currently testing AI-driven spot detection as a non signal processing approach.
Thank you for that lead!
@evartsb would you be able to share the same with our team?
@svcostes would you be available for a short meeting sometime with our IRIF team: PI @joseph.j.borg PI Gianluca and Bioinformatician Josef Borg ?
It appears that our project is along the same lines as your work with few ramifications.
Best,
Anu
Sure, always happy to meet Joseph
@anuiris I cna share the python scripts. There are issues with the update function since Jupyter Notebooks update to newest version and the screen refresh leaves some artifacts from the previous batch of images (just affecting image border and image text description.) but the script works just find. It loads 40 images into a viewer in Jupyter Lab, then you right click, or left click on an image to label it. What is your email so I can email it to you?
Also, are you interested in joining our project team bi-weekly meetings every other Monday at 2-3 PM ET ?
@evartsb Thank you. Could you please share it to Dr Josef Borg as well? josef.borg@um.edu.mt
My email ID is
anuevangelineiris@gmail.com
Yes, I’d like to join your group and attend the meetings as well. Thank you very much.
@evartsb @svcostes @rtscott2001
Is there an inclusion criteria for manually labelling an IRIF? Some images have multiple foci of varying intensities that generate confusion as to which is a true foci and which are Artifacts.
IRIF intensity cannot be used as a cutoff alone. You need to accept lower intensity foci as well. We typically use the 0 Gy control to tune detection and intensity threshold. Having said that, expect to also see foci in non-irradiated samples. They reflect real DNA DSB due to other stressors than Ionizing Radiation. But if you set intensity very low, then you also start detecting chromatin patterns, often punctuated as well. That is the reason why I typically use pattern recognition signal processing for detection. It is not as sensitive on intensity and does a great work differentiating irradiated vs non irradiated. Also, always tune the detection based on the highest dose and shortest time point.
Here a few more pointers:
Spatiotemporal characterization of ionizing radiation induced DNA damage foci and their relation to chromatin organization - ScienceDirect
Thank you for the pointers. The dataset that we are using for the project is OSDR366. We couldn’t find 0Gy control images from the public domain. Is it available? We tried LabelMe and it offsets the foci boundary quite a lot. Currently Josef has created a tool with pattern and intensity combined recognition. Once we have few images labelled well, we shall show them to you for your expert opinion.
Thank you!