Physicians spend less time than ever with patients — just 27 percent of the workday, according to a study published in the Annals of Internal Medicine in 2016.
The main culprit: electronic health records. Doctors find themselves increasingly glued to computers, acting as glorified data entry administrators.
Even when they’re in the same room as patients, doctors interacted with them only 52 percent of the time. However, the study also found a contra-indicator: doctors who used some kind of document support — a medical scribe or dictation service — spent more time interacting directly with patients.
That’s a dynamic LexiconAI hopes to capitalize on using GPU-infused AI.
“We thought this was a problem that we could tackle,” said Matt Rubashkin, co-founder and CEO of the Silicon Valley-based startup. “There really needs to be a better way to attack the system. How do we empower doctors and help them focus on what’s important?”
Recapturing Wasted Time
Rubashkin and LexiconAI co-founder and CTO Ian Plosker both had worked in the digital health area previously, and saw firsthand how much time was being wasted on documentation.
The two joined forces with the intent of leveraging voice and speech recognition to reinvent how medical data is captured. They focused on using deep learning to let providers capture medical information more seamlessly, without interrupting their patient interactions. The result: LexiconMD, a mobile app that takes in unstructured speech and spits out structured data.
The app records the conversation between doctor and patient and streams the audio to LexiconAI’s cloud-based engine, which returns the captured text —complete with best word suggestions — in just 500 milliseconds.
The app integrates with many electronic health record systems to make it possible to automatically fill the right fields with the returned data, and Rubashkin claims that LexiconMD is 94 percent accurate out of the box. (For systems with which LexiconMD isn’t yet integrated, physicians can still use the speech recognition capabilities and simply plug the data that’s returned to them into the correct fields manually.)
“When people interact with LexiconMD, it’s like interacting with a human,” said Rubashkin. “Instead of you having to use specific words and adapt to it, our goal is for LexiconMD to adapt and learn from you.”
LexiconAI’s initial focus has been on the senior care market, for which it has created models built using specific datasets based on the language used by geriatric care providers. And it has integrated tightly with PointClickCare, the top long-term care electronic health records system.
Rubashkin notes that one senior care provider, Juniper Communities, is saving significant money with LexiconMD by reducing the time nurses spend on documentation and improving the quality of that documentation.
NVIDIA GPUs figure prominently in LexiconAI’s offerings in two ways. First, the company uses a cluster of NVIDIA Tesla GPU accelerators in the Google Cloud Platform to run its service. This enables the quick turnaround of speech recognition analysis and the auto-filling of electronic health record systems.
The company also uses NVIDIA TITAN X GPUs in its office to enable experimentation without having to spin up more compute instances. It’s also using CUDA, cuDNN and TensorFlow to leverage those GPUs for deep learning experiments.
“It would’ve taken years for us to get where we are now without GPUs,” said Rubashkin.
Now, just a year into the life of LexiconAI, Rubashkin sees numerous opportunities for the company’s technology to help a variety of professionals buried under burdensome documentation. It’s already working on an app for the legal arena that it hopes to make available soon.
The company also wants to offer more real-time services to assist professionals. For instance, Rubashkin said LexiconMD’s AI-powered platform could help doctors find information and assist in medical decision making, much like in aviation, where electronic alerts and real-time display of information play a huge role for pilots.
“There are clever ways to learn from other markets where augmented intelligence has enabled dramatic improvements in productivity and safety,” he said.
LexiconAI is one of more than 2,200 startups in our Inception program. The virtual accelerator program provides startups with access to technology, expertise and marketing support.