AI's first big success in healthcare: radiology

What happened

The most comprehensive study about breast cancer screening powered by AI

The Lancet just published a study about using AI to power breast cancer screening that included a massive 105'934 participants. AI-supported screening resulted in a 29% increase in cancer detection rate compared to standard screening (6.4 vs 5.0 per 1,000 participants).

The increase was mainly in the detection of small, lymph-node negative, invasive cancers, including hard to detect ones (e.g., triple-negative, HER2-positive). The recall rate and false-positive rate were not significantly different between the groups, but the positive predictive value of recall was higher in the AI-supported group.

In the meantime, the workload for radiologists was reduced by 44.2% in the AI-supported group.

Impact on our Investment Case

Automatization is the way to decrease total healthcare expenditure

Two-thirds of the $4tn U.S. national healthcare expenditure can be disrupted by automation, whether through physical automation or, primarily, data automation. When focusing on data automation, we can categorize data into administrative and clinical segments. While administrative automation is estimated to yield $1tn in savings, even greater value could be extracted from automating clinical data.

Currently, clinical data is fragmented and requires manual collection and curation, making it difficult to extract value, such as enabling accurate diagnostics, without specialist intervention. Yet while demand has surged fivefold in the U.S. over the past decade, the number of radiologists has remained unchanged. Medical imaging is one of the most in-demand diagnostic fields, and where AI can find immediate use. We have been mentioning this since 2019: automating radiology workflows through AI is essential to address the rising demand for medical imaging.  

AI in radiology is the most significant portion of AI-approved software/devices

The FDA has now approved over 1,000 AI/ML-powered medical devices. The first 20 approvals (1995 - 2015) took 20 years, whereas 700 devices have been approved in just the last five years, 79% of which are for radiology applications. Most AI-powered medical devices were approved based on performance on known datasets rather than clinical trials. With an increasing number of studies demonstrating AI’s effectiveness in radiology, it is becoming clear that radiology is AI’s first major success story in healthcare. Only a handful of companies own the majority of the intellectual property behind these AI models. Therefore, it is crucial to carefully select models and companies that will dominate and capture market share.

Our exposure at atonra

Our Bionics strategies have maintained significantly higher exposure to AI in radiology than most of our peers. While a market-neutral strategy should have around 4% exposure in this sector, our portfolio's exposure is at ~11%. Our investments cover the entire value chain:

• Imaging Equipment: GE HealthCare Technologies Inc, which leads the market with the highest number of FDA-approved AI software—twice as many as Siemens Healthineers, its closest competitor.
• Radiology & Diagnostic Centers: Radnet Inc, providing patients and radiologists with greater access to medical imaging.
• AI Software Suites: Pro Medicus Limited, streamlining image analysis and accelerating radiology workflows.

As cancer rates increase, and orthopedic needs are driven by an aging population, the demand for medical imaging and diagnostics will continue to grow, benefiting all three segments.

Our Takeaway

We are not alone in recognizing AI's transformative potential in healthcare. Google said the next trillion-dollar business is at the intersection of tech and healthcare. Nvidia's CEO Jensen Huang said, "The single greatest potential of AI is to transform biology from a life science into life engineering to improve healthcare." Sam Altman, CEO of OpenAI, predicted, on AI's impact on medical research, " We will see diseases get cured at an unprecedented rate." 

The recent study published in The Lancet provides concrete evidence of AI's immediate, tangible benefits to radiologists. This demonstrates that delivering better care at reduced costs is achievable on an unprecedented scale, and we plan to keep an overweight exposure to this area of our investment universe.