GenAI’s impact widens the scope for AI
With the success of ChatGPT, Generative AI (#GenAI) technologies, particularly large language models (LLMs), have dominated the discussion around artificial intelligence throughout the industry. The same is true for Radiology AI, so the influence of new GenAI technologies was palpable in almost every encounter at RSNA.
First and foremost, GenAI significantly widened interest and expectations in AI within the radiology community. While in previous years, the primary driver for new products and companies came from image analysis and processing tools, in 2023, discussions around future opportunities centered around LLMs’ capabilities. This included expectations for streamlining workflows, such as automatic report generation, which could significantly simplify the integration of AI-generated results with radiologists’ assessments.
Also, regarding priority for healthcare institutions, the impact of LLMs is noticeable. The general trend of AI permeating all aspects of business and the fact that many decision-makers have now had their own experiences with impressive AI results seems to pave the way for discussions about AI with the C-level in healthcare institutions.
Incidentally, the announcements of foundation models, that is general models fundamentally characterizing imaging modalities and linking these to textual representations such as radiology reports, probably were the single most impactful news items this year. We will have to wait for the practical implications of these foundation models. Still, they represent a shift from a narrow approach, focusing on individual findings, to a more general stance towards AI-assisted reading.
Real-world value of AI increasingly in the limelight
While the application areas for AI are widening, healthcare decision-makers’ expectations on return on investment (ROI) are rising. However, as experience with AI matures, concrete evidence of proven clinical or business value remains scarce. Only for select solutions reimbursement pathways have been established, and creating generalized health economic assessments continues to be a complex endeavour.
Typically, vendors engage with providers on a case-by-case basis for ROI calculations, and the very different mechanisms of value-generation for imaging centers and hospitals require a nuanced approach from the vendors. A strong focus on demonstrating customers’ business value is a clear opportunity for AI vendors to differentiate themselves from the competition. This requires a further shift away from technology-focused thinking to holistically understanding the context of users’ workflows and the financial mechanisms for procuring enterprises.
Quite contrary to the previous discussion about AI potentially replacing radiologists, one of the most pressing issues in the profession still is staff shortage and radiologists’ burnout. Solutions addressing access to care, reducing workload or making the jobs to be done less stressful, stand a good chance of capturing the attention of decision-makers.
The cost and practicalities of deployment have also gained more attention. Indeed, widespread adoption has not yet been achieved for AI in radiology, neither in Europe (where some countries like Netherlands and France are ahead of the curve) nor in the US, but as use is increasing, the talk now is about efficient utilization of AI.
The platform game – increasing number of deployment models
As in previous years, many discussions centered on delivery channels. It remains to be seen what the dominant access pathway to AI functionalities will be: dedicated AI delivery platforms such as Calantic / Blackford, INCEPTO, deepc, CARPL – Radiology AI Platform or Eureka Clinical AI | Formerly known as EnvoyAI; legacy IT vendors such as the PACS and Enterprise Imaging companies; or flagship AI vendors with a larger client base and substantial funding such as Aidoc or Viz.ai, which have been complementing their application portfolios through partnerships with other AI vendors.
A recurring theme in many presentations, as well as from various vendors on the show floor, was the validation and monitoring of solutions. Be it from the requirements of post-market surveillance, the impact of device selection or parameters on the performance of tools, or changes in patient cohorts, the need to evaluate tools on providers’ data and continuous monitoring of performance now seems to be a given. However, this adds another layer of complexity to the routine adoption of AI that client-facing vendors need to accommodate.
Focus shifts from narrow diagnostic tools to clinical pathway solutions
From very different angles, we have noticed a converging trend towards broader AI-support for clinical pathways. We have already covered the expectation on wider radiological workflow support through LLMs above, but also from the perspective of the clinical patient management and care pathways, the role for AI solutions is widening. Instead of viewing AI as a product by itself, AI is increasingly perceived as a facilitator of improved care delivery.
This change of perception is captured, for example, in breast cancer screening, where it has become clear that vendors cannot just provide a “better CAD” solution but need to demonstrate the concrete realization in real-world screening settings (for different countries or even for different risk groups), in which their tools will prove tangible value. Vendors’ offerings will therefore likely transition from 2nd reader functionalities to AI-driven screening management solutions.
In general, use cases in which AI can influence the patient journey or the healthcare systems’ cost or quality, for example, by reducing length-of-stay or by enabling appropriate procedures, have the potential to tap into much larger financial benefits. This also includes crossing the departmental silos between radiology and other clinical disciplines, such as oncology, to create an integrated data-driven view of a patient. Such disease- or patient-centric solutions hold tremendous potential for clinical care, though certainly complex challenges remain. Companies such as Tempus are at the forefront of this change.
Bright outlook from a bird’s-eye-view on the AI industry
The picture is complex regarding the general state of the radiology AI scene, particularly the startup landscape. Expectations of consolidation in the market have been widespread for years, but the actual market dynamics have shown only a handful of recent transactions. Consequently, the AI showcase once again boasted an overwhelming number of companies competing for attention. While some of the larger vendors have notably decreased their booth sizes and a few companies ceased their presence altogether, overall, the AI sector’s footprint in the hall still by far exceeded the designated AI showcase area.
The tech funding situation, in general, remains challenging, certainly for smaller startups, and several of our conversations indicate that the market appears to be splitting up between well-funded global leaders and smaller vendors now shifting their focus towards near-term profitability, hinting at a potential coexistence and eventual consolidation.
Such a shift towards profitability can mean a focus on specific funding mechanisms in vendors’ home markets, downsizing sales capacity while partnering with companies that provide customer access or turning towards income mechanisms outside of the realm of diagnostic radiology.
In conclusion, RSNA 2023 has provided a comprehensive snapshot of the evolving landscape of radiology AI, marked by a maturing industry, diverse delivery models, and generally an optimistic outlook toward the future, rekindled by the promises of LLMs and GenAI.
Connect with us to engage in this discussion
As we navigate these developments, we are committed to staying at the forefront of advancements and leveraging these insights to benefit our customers. Helping organizations position themselves in the complex field of deep tech in healthcare and focus their strategy on future business value, we are keen on exchanging thoughts about the direction of AI in healthcare and the clinical impact novel technological innovation can achieve.
What are your main takeaways from RSNA 2023? Do you agree with our perspective?
Yours,
Dr. Stefan Braunewell
Managing Partner, Synwisery
