BrainBot

In CT scans, acute hemorrhages appear bright. Radiologists identify subtypes based on location, shape, and proximity to other structures.
Intraparenchymal are in the brain, intraventricular and subarachnoid in fluid spaces. Extra-axial hemorrhages collect outside the tissue.

BrainBot stands for Intracranial Hemorrhage Detection Robot and it is a free-access diagnosis tool for classifying intracranial hemorrhage. BrainBot is based on a Vision Transformer (ViT). Our purpose is to provide a free, easy-to-use, anonymous and fast tool to give physicians a first diagnosis, as well as a website where useful and up-to-date information can be gathered.

Face to medical professionals, neurosurgical experts, particularly radiologists in diagnosing intracranial hemorrhages

Why BrainBot

Rapid identification of acute intracranial hemorrhages and subtypes
Single-image classification capability
Feature for doctors to upload professional assessments
Collection of labeled data for future system improvement

We are proud of this holistic approach that addresses both immediate diagnostic needs and long-term system optimization.

Inspiration

Intracranial hemorrhage is a critical medical issue, representing 10% of strokes in the U.S., where stroke is a leading cause of death. Rapid and precise diagnosis is essential and typically involves complex medical imaging. Gathering annotations for large medical datasets poses significant challenges for developing deep learning(DL) systems in healthcare. Main issues include:

Privacy concerns and the need for expert manual annotation, making the process costly and impractical.
Time-consuming, the problem intensifies for tasks like lesion segmentation in 3D CT scans, where a single annotation can take an expert over 5 minutes.

Specifically, DL enhances analysis in head CT scans, which are the go-to diagnostic method given their quickness and accessibility. This technological shift is pivotal for helping radiologists make faster, more accurate decisions.

How we built it

The model was trained based on the RSNA Intracranial Hemorrhage Detection data set, available at Kaggle. The website was built with HTML and CSS using the Bootstrap framework so it would be responsive. Finally, integration of the Transformer model with the website was performed with the Flask framework.

Challenges we ran into

One of the key challenges we are facing is the overwhelming size of our dataset, which makes user interaction difficult due to the absence of adequate server resources. Despite the availability of code references from previous work, the limited time frame exacerbates the issue, making it challenging to perform predictions and computations for even a single image.

What's next for BrainBot

Some features that could be added in future versions include:

Quantify the volume of intracranial hemorrhage and assess whether the brain's midline axis is shifted
Train the model with a larger dataset to improve accuracy
Implement image segmentation techniques to directly visualize the location of bleeding