Healthcare AI companies claim widespread deployment whilst clinical adoption lags
Corporate narratives about "operational infrastructure" in neurology mask pilot programmes and fragile partnerships
When icometrix announced its latest partnership with Philips in November 2024, the Belgian AI company's press release confidently declared that artificial intelligence in brain imaging had moved beyond experimental research to become "operational infrastructure." The firm claimed deployment across "over 300 hospitals worldwide" with "reimbursement channels in place."
Yet six months earlier, surveys revealed that 61% of UK neurologists and radiologists had never used AI in clinical practice. Something wasn't adding up.
This disconnect illuminates a broader pattern across healthcare AI, companies routinely present early-stage pilots and partnership announcements as evidence of widespread clinical deployment, creating a mirage of adoption that obscures the genuine challenges preventing AI from reaching patients at scale.
The gap matters because it shapes billions in investment decisions, influences policy frameworks, and ultimately determines whether promising technologies reach the patients who need them. When marketing narratives outpace clinical reality, everyone loses.
The deployment illusion
Icometrix's trajectory reveals how companies construct deployment narratives. The firm celebrated "over 300 hospitals" using its icobrain platform whilst simultaneously announcing major partnerships with Philips and Siemens that would integrate their technology into MRI systems for the first time.
The timing suggests that previous "deployments" may have been limited pilots rather than the operational integration implied by their marketing. Baptist Health's October 2024 announcement that it was becoming "the first health care organisation in Arkansas to utilise icometrix's cutting-edge AI technology" raises further questions about the depth of those 300+ claimed deployments.
This pattern repeats across the sector. Companies cite hospital counts, FDA clearances, and partnership announcements as evidence of clinical transformation whilst the underlying adoption remains shallow. The result is an industry narrative that consistently overstates current reality whilst understating remaining challenges.
Reimbursement theatre
The payment landscape exposes similar sleight of hand. Icometrix trumpets CPT codes 0865T and 0866T as proof of "reimbursement channels in place," but these Category III codes became active only in January 2024 and guarantee nothing. They're temporary designations for emerging technologies, designed for data collection rather than payment assurance.
Consider the numbers, icometrix claims "over 30 insurers" now reimburse for AI brain analysis. That sounds impressive until you realise thousands of insurance entities operate across the United States. Medicare and Medicaid—covering 140 million Americans—don't automatically reimburse Category III codes, leaving coverage to local administrator discretion.
Even successful reimbursement stories reveal the precarious nature of current arrangements. Viz.ai achieved Medicare reimbursement for stroke detection in 2020, but such examples remain exceptional. Most AI applications continue operating in reimbursement limbo that prevents sustainable deployment at scale.
The clinical reality check
The most telling evidence comes from practising neurologists themselves. Dr Bing Yu Chen at Cleveland Clinic articulates the core problem, "A major concern is the 'black box' nature of AI algorithms—they often lack transparency, making it difficult for clinicians to understand the logic or data underlying their predictions. This opacity can undermine trust, especially when patient lives are at stake."
Workflow integration proves equally challenging. AI tools must seamlessly connect with electronic health records, imaging systems, and clinical processes developed over decades. Technical incompatibilities and training requirements often derail implementation beyond initial pilot phases.
A 2025 survey of UK clinicians found that 71% preferred AI-assisted triage for brain MRI, yet the majority had never used such tools clinically. The enthusiasm exists, but the practical barriers remain formidable.
The partnership pathway deception
Understanding how companies achieve their deployment numbers reveals a crucial distinction. Rather than hospitals independently purchasing AI software, companies increasingly rely on partnerships with equipment manufacturers who embed AI tools within existing systems.
Icometrix's Philips partnership exemplifies this model. Instead of 300 separate hospital decisions to adopt AI, the technology becomes part of BlueSeal MR scanners and imaging platforms. Similarly, their Siemens collaboration makes tools available through digital marketplaces to existing customers.
This approach can rapidly multiply access points whilst obscuring actual usage. "Hospital deployment" may reflect potential access rather than active clinical integration. The difference between having AI tools available and actually using them routinely represents a chasm that marketing materials rarely acknowledge.
Evidence gaps persist
Clinical validation presents another layer of complexity. Some studies demonstrate genuine improvements—icometrix's AI achieved 93.3% sensitivity for detecting multiple sclerosis activity compared to 58.3% for standard radiology reports in a study published in Nature Digital Medicine.
Yet independent, long-term outcome data remains scarce. Much evidence comes from company-sponsored studies or short-term pilots. The FDA's recent clearance of icometrix's ARIA monitoring system represents legitimate regulatory validation, but focuses on safety monitoring rather than diagnostic improvement or patient outcomes.
Healthcare demands evidence that AI actually improves patient lives, not just technical metrics. That proof remains largely absent across neurology AI applications.
Market incentives driving distortion
Venture capital dynamics encourage this reality distortion. Investors reward rapid growth metrics, pushing companies to emphasise deployment numbers and partnership announcements. Healthcare systems prioritise patient safety and workflow integration—requirements favouring gradual adoption timelines.
The result is systematic overselling of current capabilities. "Operational infrastructure" describes what might more accurately be characterised as promising but unproven pilots. Companies face genuine pressure to show progress, but presenting aspiration as accomplishment serves no one's long-term interests.
The cost of overselling
This gap between claims and reality carries real consequences. Investors make decisions based on inflated adoption metrics. Policymakers develop frameworks assuming capabilities that don't yet exist at scale. Healthcare providers face unrealistic expectations about AI's immediate impact.
Most critically, patients miss out on genuinely valuable applications because overselling undermines trust whilst creating impossible standards for emerging technologies.
The technology itself shows legitimate promise. AI in neurology has demonstrated improved diagnostic accuracy and offers solutions to workforce shortages. But realising this potential requires honest assessment of current limitations rather than marketing fantasy.
A path forward
Moving beyond deployment theatre requires fundamental changes. Companies must distinguish clearly between pilot programmes and operational integration. Reimbursement pathways need development beyond temporary Category III codes. Most critically, the clinical evidence base requires expansion through independent outcome studies.
Healthcare AI may indeed transform neurology practice, but transformation will prove more gradual and complex than current narratives suggest. Success requires acknowledging this reality rather than perpetuating unrealistic expectations.
The revolution may be coming, but it hasn't arrived yet. Recognising that truth offers the best foundation for sustainable progress toward AI's genuine potential in patient care.