How GE Healthcare is using data science and AI smarter, CIO News, ET CIO
Medical errors are the third leading cause of death worldwide, after cardiovascular disease and cancer. Thus, GE Healthcare uses technologies such as artificial intelligence (AI) and data science to identify ways to reduce the potential for medical risks in healthcare.
According to Girish Raghavan, vice president of digital engineering and technology at GE Healthcare, the most pressing issue for healthcare today is data. While most healthcare organizations aim to enter precision medicine or predictive health, data becomes the heart of predictive health in precision medicine.
“For example, if you go to a hospital, you have a pathology department where you do a lab test. Then you go to an imaging department and do an X-ray, CT scan, or MRI. Then you go to a doctor. He wouldn’t have any prescriptions written, and they were probably written three months ago by another doctor because you have a history. Then we have a dynamic with the whole Covid scenario. You would have heard of genome sequencing where you want to see which variant of COVID is affecting patients, and that requires a lot of data and predictions,” he said.
With data distributed everywhere, the challenge is to find a process to bring the data together and get useful insights. GE Healthcare is at the forefront of research in this area and uses data science to find ways to integrate data from disparate systems across hospitals to gain valuable insights while using the appropriate advanced technology.
First data quality, then AI
“I think as an industry grappling with this data, AI as a technology hasn’t matured yet, but our data is good because in healthcare, tolerance for error is very minimal, and one error leads to misdiagnosis. We are extremely cautious. The healthcare industry has not exploited or fully utilized AI and machine learning. We are still a long way off because we have to wrong with that. Data science is a huge field for us. Today, our challenge in the industry is that data quality is an issue. Data quality is still suboptimal,” he said. he declares.
Around the world, as demographics become a significant frontier contributor to the entire environment, the company has pockets of data around the world. Raghavan says the challenge is to get data from across the ecosystem while maintaining good data quality.
“Because, for example, you take data from the United States, hypothetically, you build a model based on the American population. If you take the same model and apply it to an Indian population, it may not work. So demographic differences have a huge impact on us. The data quality, the demographic differences, the sample data we have when we continue to use the highest number of records, there will be a 1% improvement, but at some point it stops” , explained Raghavan.
He says the focus is on how to ensure that false positives are few or false and negatives are minimal. So, as a concept and technology, the company uses convolutional networks, which are mature.
“But, today, it’s just that when you combine these advanced technologies with data, the combination that will deliver the precise outcome is where we struggle as an industry,” he added.
Solve the problem
Therefore, the organization prioritizes data integrity and makes extensive use of natural language processing (NLP).
“If you look at natural language processing, it’s a very important tool that we use today to convert unstructured data into structured data. We’ve built a fair amount of skill on NLP, which will be able to convert automatically unstructured data into structured data. That comes on the one hand, but when it comes to the quality of the data itself, what we do as a company is hire some of the best clinicians across the world,” he added.
Therefore, all data used by the company is processed through a cleaning process. Before starting to implement it in the space, the company cleans, infers, and shapes the data, annotates it, and stores it for further processing in a way that maintains common sense of purpose.
“Because our model is only as good as the data, if the data is bad the whole system is broken. So we have a system in place where clinicians will look at a chart. We have several checks and balances in place to ensure that the quality of the data is not compromised so the output is also solid We are working on both fronts, both in using NLP very effectively to go from standard to another then from unstructured to structured mechanism,” he continued.
Additionally, the company works with some of the major institutes like Stanford and collaborates in research to refine the intelligence of NLP.
“Each year we generate something like 2 billion scans on GE equipment. We know roughly how things are spread across multiple countries and what variations we have from country to country. This is our biggest advantage. We have the opportunity to work with key leaders in those countries to get that information inside to make us much more robust,” Raghavan added.
Another advantage is the number of clinical experts as they have a battery of clinical experts from different countries. “I think some of the best doctors in our network can give advice and suggestions on how to move forward with this. This is how we have an advantage, and we will continue to grow in the future,” he said.