to Create Intelligent Surgical Tools
Bringing artificial intelligence (AI) into the operating room has the potential to save lives, improve safety, and produce better patient outcomes. Building intelligent surgical tools requires two types of expertise: the technical capacity to create AI tools and medical expertise to label large quantities of surgical images for training the AI model accurately.
However, creating these accurately labeled large datasets requires data from multiple healthcare institutions and time-intensive annotation by medical experts. Unfortunately, many AI companies lack access to data and the medical expertise required to understand the surgical images or label objects of interest like tumors, blood vessels, organs, etc.
Similarly, many healthcare institutions have surgical data and access to medical expertise but not the time to annotate or sufficient technical capacity to build AI models. Others have the time for annotation but not access to large repositories of data. For these reasons, collaborations are essential. There are, however, many hurdles to collaboration, including regulatory obstacles to data sharing, questions of data ownership, and annotation tools that do not support the delegation of tasks.
Our Solution
RediMinds has purpose-built a secure platform, “Ground Truth Factory” (GTF), for collaborative development of AI surgical tools. This platform supports data upload, annotation, management, and model building. It is designed to enable secure, fast, and efficient collaboration while reducing the time necessary for annotation.
Our platform is secure and HIPAA compliant. We put a process in place that helps users anonymize the data before uploading it to the platform. In addition, data storage, access, and processing services are governed and compliant with HIPAA guidelines and more.
The entire environment sits behind a firewall that protects it from DDoS, Password attacks, and hacking attempts. Data can be stored in the country where it was created, facilitating compliance with data localization requirements.
Even though institutions can share access to data for annotation, they retain control and ownership of their own data and can select how their data will be used. The platform makes it easy to assign annotation jobs to approved experts who are willing to participate.
No Test
Our team has also created an AI assistive tool to help surgeons skip through irrelevant images. This tool curates sets of images with relevant landmarks; for example, it can select a collection of images containing spleens using a baseline deep-learning model. This protects the valuable time of surgeons who annotate images.
Three geographically distributed urologists annotated over 3000 surgical images from 5 surgeries for 3 critical objects to test the annotation tool. This work has already created a preliminary model that can accurately identify the spleen, renal vein, renal artery, and background in laparoscopic donor nephrectomy videos. As our model is expanded to identify more structures, it will help surgeons navigate through dense vasculature in real-time.
The key benefits of this platform are:
- Secure and HIPAA compliant platform
- Facilitates collaboration with approved participants
- Speedy Annotation Process
- Allows access to RediMinds AI expertise for AI development
RediMinds is also adding a CT annotation tool to this platform. This will allow researchers to collaborate and annotate CT scans.