The past few years have seen a surge in demand for healthcare AI. This doesn’t show any signs of stopping soon — with three-quarters of the country’s providers and payers reporting that they have increased their IT spending over the past year.
The healthcare sector’s enthusiasm for AI has given way to hundreds of startups selling AI-powered products designed for providers, payers and biotech companies — and these AI startups must prove their value to investors in order to secure venture capital.
In the future, healthcare AI startups that offer multimodal models may have an easier time winning over investors than healthcare AI companies that don’t, according to a recent report released by Bessemer Venture Partners.
Until recently, most healthcare AI models were designed for a singular data type, such as audio, video, medical imaging, clinical records or wearable device data, the report noted.
However, efficient healthcare delivery often requires multi-dimensional data. Multimodal AI models are attractive to investors because they can operate on a wide range of data modalities and applications, the report stated.
“Change happens slowly, and then all at once. While it’s understandable that healthcare executives aren’t yet championing multi-modal AI, given its nascent status and still-developing applications, this technology deserves greater focus as research translates into products. We’ve recently witnessed a similar transition with large language models, which have rapidly moved from research to widespread application,” said Morgan Cheatham, vice president at Bessemer.
Multimodal AI differs from models trained on single data types — such as language models specialized in text — because it is uniquely positioned to capture “the rich, multifaceted nature” of healthcare and biomedicine, he explained.
In his view, multimodal AI models are better suited to collect and leverage relevant data — which could be related to clinical events, imaging, operations, access, social determinants of health or patient-reported outcomes.
“By integrating diverse types of data, multimodal AI has the potential to provide more comprehensive insights and solutions in the complex healthcare landscape,” Cheatham declared. “We anticipate the next few years will usher in a renaissance for multimodal AI in healthcare, with an application space surpassing anything we’ve seen before.”
Greater data integration comes with additional risks, though. The rise of healthcare AI introduces new concerns when it comes to data privacy and security, Cheatham noted.
Regulatory and legal concerns remain a top barrier to AI adoption among healthcare organizations, with 43% of providers and 38% of payers citing these risks as hurdles, he pointed out.
“As with previous technological shifts in healthcare, the success of AI will depend on both technological advancements and regulatory progress. It will be critical to integrate insights and address concerns from a diverse array of stakeholders, ranging from startups to Fortune 500 companies, and from early-stage researchers to healthcare executives,” he declared.
Photo: metamorworks, Getty Images
