Biomedical Engineering & Technology

Abstract

Healthcare systems worldwide are facing a critical challenge with a shrinking pool of doctors and a rapidly growing and aging 
patient population. LLMs and other self-supervised models can change the situation but require access to a large volume of data to 
be generalizable and less biased. Without large amount of data, the models are not generalizable and biased. ChatGPT is trained on 
publicly available internet data, and even they report that it is no longer enough.

Imagine how dire the situation is in AI in Healthcare that requires access to Personal Medical records. These records are not just bits 
and bytes, but somebody’s very personal very private data. E.g., on the dark web price for a person’s healthcare record is 500x the 
price for a credit report. To give AI in Healthcare access to the personal medical data, the data must be deidentified. 
Local large volume deidentified multimodal medical data lakes offer a feasible solution that empowers hospitals and individual 
patients to make a significant impact, while monetizing their data.