Accurate Cell Labeling for AI: Why Medical Students are Needed

SuperAnnotate needed quality instance segmentation performed on hundreds of cells to train their artificial intelligence. Instance segmentation is a type of computer vision used on medical images to identify and segment individual cells. Annotating cells is critical for analyzing the properties of individual cells and understanding their role in various medical conditions. While instance segmentation is a complex and time-consuming task, medical students can be a valuable resource for performing this task. In this blog post, we will explore how we created a team of medical students to train Superannotate’s AI. 

‍

For this project, we received a dataset of microscopy images with specific cell types pre-indicated by a marker. SuperAnnotate trained a machine learning model from external data to perform initial instance segmentation. Due to inaccuracy of initial labeling by their ML model, Yenda's responsibility was to fix the errors made by the AI by editing the annotations. Along with editing, our team created new annotations around cells that the AI initially missed. 

‍

We partnered with the University of Zambia’s School of Medicine to connect their students to this remote earning opportunity.

‍

To create the team, we partnered with University of Zambia’s School of Medicine to get a list of their top performing students. Once we interviewed all candidates interested, we created a cohort in Google classroom to train the group on the project. Lastly, SuperAnnotate provided training on the specifics of their tool. Training ML models in-house is time consuming and expensive. The crowd is the cheapest option but the quality is low and there is a lack of visibility and management. Partnering with medical students is a great way to utilize their medical knowledge, provide great work experience, and save money compared to in-house doctors training the data.

‍

Here are three reasons why SuperAnnotate chose to use a Yenda team to train their medical AI: 

1. Medical expertise is necessary when it comes to identifying and segmenting individual cells for medical images. Our medical students understand the structure and function of cells and the various cell types.
2. Medical students are highly motivated to learn and contribute to medical AI and advancements in medicine. Instance segmentation is a critical task that is often time consuming and requires attention to detail. 
3. Medical students are familiar with medical research protocols and procedures which trains them to ensure data is collected and analyzed accurately; This is especially useful to train AI models, particularly instance segmentation on cells which requires a consistent and standardized approach.

‍

We partner with universities in Zambia to connect you to industry-specific talent. Teaming up with Zambia’s medical students is a great way to guarantee quality labeled data because they have a strong foundation of medical knowledge, are motivated to learn and contribute to medicine, and have critical attention to detail for data accuracy.

‍