Problem and Motivation
Started with an aim to build a skin disease detection tool
Realized that securing essential building blocks are a big challenge!
Data Collection
No data exists for our target population and problem
Quality Control
Data annotation happens in iterations that are time consuming and of varied quality
There is a need for skilled annotators with medical training
Small Title
Small Title
High quality datasets are then channeled to building successful AI Applications
Enabling Technologies
Clara AI-Assisted Annotation
Enable medical viewers to be AI-powered and speed-up creation of high-quality annotated datasets
Web Application Frontend and API
User registration / workflow for validation of scientific credentials
Dataset management / Visualizations / Reports
Billing / Payment
Deep Learning framework specific for medical domain use cases
Optimized for multi-GPU data parallelism
Distributed training of models without compromising privacy of patient data
Powered by NVIDIA Clara train SDK
Federated & Transfer Learning
High Availability GPU accelerated compute
S3 Secure & Efficient Dataset storage
FOSS toolkit for interactive medical image processing software, leveraging support for NVIDIA Clara AIAA.
Secure cloud data storage using AWS S3
Strict Dataset User Access management
Federated Learning solutions for private datasets
Security
Leveraging NVIDIA Clara BYOM and AIAA, Active Learning pipeline allows continuous edge cases detection, labelling and retraining of models to achieve better accuracy
Active Learning