Annotation of skeletal muscle cells
Must-read Case Studies
Foodtech App
3D Point Cloud Annotation Tool
Aerial images
Medical 3D annotation
Microscopic annotation
...and many others case studies that are confidential!
Mitochondrial diseases are individually uncommon but are collectively the most common metabolic disorder affecting 1 in 5,000 people. They can cause severe disabilities and adversely affect the life expectancy of patients.
Analysis of these images have required semi-automated annotation of thousands of Skeletal muscles in IMC images of patient muscle biopsies.
Context
CONTEXT
University of Newcastle used Explainable AI to make progress on mitochondrial diseases research.
Screenshot taken on Zeiss Arvis. The image taken by mass cytometry contains 3148 muscle fibers.
To understand mitochondrial diseases, the first step is to differentiate and profile all the existing diseases. This is what Newcastle wanted to do by analyzing images of Skeletal muscle fiber obtained by IMC (Imaging Mass Cytometry).
To train the algorithm, we need IMC images on which sick cells are annotated with a pixel-perfect precision.
To carry out this annotation project, Newcastle asked People for AI to scale this very specific task.
OUR SOLUTION
OUR SOLUTION
For this expert task, we set up close communication with the client via emails and Q&A file.
We used the annotation tool recommended by the client, ZEISS Arivis [arivis.com]. This tool is especially adapted to label microscopic data.
We created an edgecase class for the experts to review the most difficult cell labeled.
Absence of Annotation Guidelines was a big challenge at the beginning. However, by asking many questions to the client, we finally clarified the task!
Weekly meetings were also planned to address potential issues and edge-cases.
We labeled on average one cell per minute with a pixel-perfect segmentation and there are up to 10.000 cells per image!
We have selected and trained 10 labelers and a project manager. The selection was based on the successful experience from previous projects using microscopic images.
Communication and organization was key as sometimes, experts did not agree with each other.
Matthieu Warnier
Data Labeling Director @ People for AI
« At People For AI, we usually work for companies that keep the label data for their own use and often don’t communicate on that work. On the contrary, Newcastle university’s plan was to make the produced label data available for everyone. It was great to be involved in such a project with a very significant impact.»
OUR IMPACT
OUR IMPACT
The researchers published the dataset we annotated in a research article.
A total of 5 person-months have been completed during the whole annotation project.
Little by little, mitochondrial diseases are better understood thanks to these advancements.
Even projects with a reduced number of classes can have so many edge cases!
Compare your estimations with the reality as the first estimates can be too optimistic.
Establish a single point of contact on the client’s side, as divergences can be costly in terms of time.
Soon after a strict selection of our labelers, almost all our methodology and our processes were used in this project to improve the skills and the final quality of the labeled data. This project was quite challenging for our team. The required expertise was finally reached within 6 weeks after the beginning of the project and the client has been very satisfied since then.
Atif Khan
PhD Researcher : AI and Data Science for Personalized Medicine Client of People For AI in 2022
« People for AI is a team of efficient professionals who deliver fast and accurate annotation work. Their ability to keep the same team from one annotation campaign to the next contributes greatly to the quality of the work produced.”
For all your projects, whether they are research-based or operational in computer vision, People for AI deploys qualified teams, recognized expertise, and tailored tools to meet each client’s requirements. Our adaptability and experience across various industries are the keys to our success.
They trust us
Our labeled data will exceed your expectations.
We provide training data. We do not provide your data to anyone else and we will not subscribe you to any newsletter without your consent.
