AI already augments clinicians today through imaging support, automated measurements, and administrative automation, and the practical next step for most health organizations is a governed pilot with built-in clinical validation and lifecycle monitoring. Two constraints shape how fast that happens: regulatory compliance across overlapping EU frameworks and the state of local data and interoperability. Neither is a reason to wait.
TL;DR:
- Most healthcare AI is currently used to support radiologists, automate administrative tasks, and aid in surgical navigation, with adoption limited by governance issues.
- Effective implementation depends on validated, narrow use cases, careful workflow integration, and proactive governance including lifecycle monitoring and staff training.
- Regulatory frameworks in the EU, such as the AI Act and MDR/IVDR, require coordinated assessments and continuous oversight, with national sandboxes like Poland’s offering pilot testing pathways.
- The main challenges for successful deployment include data bias, workflow fit, cybersecurity, and unclear procurement processes, not the technology itself.
- Progress toward scaling AI in healthcare will focus on validated, specific tools and improved governance infrastructure, rather than broader or fully autonomous systems.
Where AI Is Already Used in Healthcare Today
AI in healthcare has moved past the demo stage in specific, narrow tasks. Imaging is the clearest example: automated measurement tools now support radiologists by flagging findings and standardizing protocol adherence, which can shorten study time compared with fully manual workflows, according to a Polish deputy health minister’s account of the Platforma Usług Inteligentnych (PUI). Chatbots and administrative tools have followed a similar path, handling scheduling and documentation drafts rather than clinical judgment.
Adoption is uneven across the region, and the gap is mostly organizational rather than technical.
- Only 4 of 50 Member States in the WHO European Region have a health-specific national AI strategy, and fewer than one in four offer formal AI training for health workers, per WHO regional reporting.
- 86% of surveyed countries cite legal uncertainty and 78% cite financial constraints as major barriers to scaling AI in health systems, the same WHO regional survey found, underscoring that governance readiness lags the technology itself.
- Imaging AI and administrative chatbots remain the two most mature categories in day-to-day practice.
Research is also pushing toward more specific, patient-adapted tools, a trajectory worth watching as pilots mature into procurement decisions.
Clinical Use Cases That Save Time and Improve Care
The highest-value use cases share a pattern: AI handles a narrow, repeatable task while a clinician retains judgment over the outcome. That framing matters both for safety and for how these tools get regulated.
- Imaging and radiology: automated measurements, triage prioritization, and protocol enforcement reduce variability and can shorten reporting time, building on the PUI example described above.
- Administrative automation: scheduling, clinical documentation drafts, and coding assistance free clinician hours for direct patient contact rather than paperwork.
- Surgical navigation: MIT researchers recently published xvr, a patient-specific method that aligns intraoperative X-rays with preoperative 3D scans quickly and with sub-millimeter precision, a signal of where procedure-assistance tools are headed.
- Remote monitoring and telemedicine: continuous data collection paired with alerting logic extends follow-up care beyond scheduled visits, particularly useful for chronic disease management.
Each use case still depends on people using it correctly, which is where deployments most often succeed or fail.
Pro Tip: Pilot one use case at a time and measure clinician time saved directly, rather than layering several AI tools into a single workflow before you know which one actually helps.
What Clinicians and Managers Must Know About AI Regulation
Regulatory clarity is improving, but it now spans multiple frameworks that apply at once rather than in sequence. Understanding how they interact is the first governance task for any team planning a deployment.
- The AI Act (Regulation (EU) 2024/1689) classifies many clinical AI tools as high-risk systems, which brings obligations on providers around risk management, data quality, and transparency.
- For medical AI, MDCG guidance from the European Commission confirms that MDR/IVDR and the AI Act apply together, requiring coordinated conformity assessments and continuous post-market surveillance rather than a single pre-market check.
- The European Health Data Space FAQ sets interoperability expectations for electronic health records that any AI tool touching patient data will need to meet.
- Poland’s own act on AI systems, in force since July 2026, establishes the KRiBSI oversight body and regulatory sandboxes, a structured path for testing pilots under supervision instead of guessing at compliance alone.
Legal uncertainty is cited by 86% of countries in the WHO European Region as a major barrier to AI adoption, so a sandbox pathway is not bureaucratic overhead. It is a direct way to reduce that specific risk, according to WHO’s regional findings.
Operational governance follows from these rules rather than sitting apart from them: lifecycle monitoring, decision documentation, explainability for clinical users, and demonstrable staff training all show up as expectations across the frameworks above. Our AI governance framework walks through how to structure that monitoring in practice.
Implementation Challenges Worth Planning for Early
Most AI deployment failures in health settings trace back to a handful of predictable causes, and none of them are exotic.
- Data representativeness and bias: a model trained on one population or one hospital’s imaging equipment often needs local fine-tuning and revalidation before it performs reliably elsewhere.
- Human-AI interface fit: poor workflow integration causes more real-world failures than weak model accuracy, since clinicians will route around a tool that adds friction instead of removing it.
- Cybersecurity and privacy: any tool touching patient records has to align with EHDS interoperability expectations and general data protection obligations, not just the AI-specific rules.
- Procurement and financing: unclear regulatory status inflates both vendor pricing and internal approval timelines, which WHO’s regional survey flags as a cost driver alongside how AI transforms compliance for security questionnaires, reducing administrative and procurement barriers.
Pro Tip: Ask any vendor to show you their post-market monitoring plan before you sign, not after deployment. If they cannot describe how they will track performance drift, that is your answer.
None of these risks argue against adoption. They argue for sequencing: validate on your own data, design for the clinicians who will actually use the tool, and build security and compliance into the architecture rather than bolting it on later.
A Roadmap From Pilot to Scale
A governed pilot moves through a predictable sequence, and skipping steps is where most projects lose time rather than save it.
- Define the clinical objective, success metric, and ethical guardrails before selecting any tool or vendor.
- Choose a constrained pilot with clinician co-design from day one and a small number of measurable endpoints.
- Prepare the regulatory checklist: conformity documentation, data protection impact assessment, and informed consent procedures where patient data is involved.
- Build a clinical validation plan with defined post-market monitoring metrics, not a one-time accuracy check.
- Plan for scale by confirming interoperability with existing EHR systems, staff training capacity, and procurement criteria before expanding beyond the pilot site.
| Roadmap stage | Primary output | Who owns it |
|---|---|---|
| Define objective | Success metrics and guardrails | Clinical leadership |
| Constrained pilot | Working prototype with clinician feedback | Clinical team and engineering |
| Regulatory checklist | Conformity documentation, consent process | Compliance and legal |
| Validation plan | Monitoring metrics, drift thresholds | Clinical and data teams |
| Scale | Interoperability and training plan | Health system management |
Our guidance for health system leaders covers the engineering trade-offs behind each of these stages in more depth.
Lessons From Building a Secure Telemedicine Platform
A project to build a secure, compliant telemedicine platform for mental health care surfaced lessons applicable to many clinical AI projects. Secure data pipelines and clear EHR integration boundaries mattered more than model sophistication. Lifecycle testing and ongoing monitoring caught issues that pre-launch testing alone would have missed.
- Design data pipelines for auditability from the first sprint, not as a retrofit before a compliance review.
- Treat EHR integration as its own workstream, with its own timeline and its own testing plan.
- Build monitoring into the platform itself so performance drift is visible without a manual audit.
When assessing a vendor or partner for similar work, ask for specifics on these three points rather than general claims of AI expertise.
What AI Means for the Healthcare Workforce
AI is changing the shape of clinical and administrative roles more than it is eliminating them outright. Radiologists working alongside automated measurement tools still make the diagnostic call, but the tasks around that call, drafting reports, flagging cases for priority review, shift toward the machine. Administrative staff see a similar pattern: scheduling and documentation drafting move to automated tools, while judgment calls about patient communication and exceptions stay with people.
The bigger workforce gap is training, not displacement. WHO reporting found that fewer than one in four countries in the European Region provide formal AI training for health workers, which means most clinicians are learning to work alongside these tools informally, on the job, without structured guidance on their limits.
That gap has real consequences. A clinician who does not understand why a model flagged a case, or why it missed one, cannot exercise the oversight the regulatory frameworks assume they will provide. Explainability requirements under the AI Act exist partly for this reason: a tool that cannot be understood by the person using it cannot be safely overseen by that person either.
Health systems that invest in structured AI literacy programs, not just tool rollout, tend to see fewer workflow breakdowns and less clinician resistance. The workforce question is less about which jobs disappear and more about which skills every clinical role now requires.
Personalization and Precision Medicine
AI’s role in precision medicine is to make personalization operationally practical at a scale manual review cannot match. Instead of applying a single protocol to every patient with a given diagnosis, models can incorporate imaging data, prior history, and treatment response patterns to suggest adjustments specific to that patient.
The xvr surgical navigation method is one illustration of this shift: by aligning intraoperative imaging with a patient’s own preoperative scans in real time, it adapts the procedure to that patient’s anatomy rather than relying on generalized reference points, according to MIT’s reporting on the technique. The same logic extends to treatment planning in oncology and chronic disease management, where models trained on relevant patient populations can help clinicians weigh options faster.
The limiting factor is data quality, not model design. A personalization tool trained on a narrow or unrepresentative population will personalize poorly, sometimes in ways that are hard to detect until a specific patient group is underserved. That is why local validation before relying on any precision medicine tool matters as much as the underlying model. Personalization built on solid, representative data is a genuine clinical advance. Personalization built on convenient data is a liability wearing a confident interface.
Where Healthcare AI Is Headed Next
The near-term trajectory in healthcare AI points toward tools that are narrower, better validated, and more tightly integrated into existing clinical systems, rather than broader or more autonomous. Surgical navigation research like xvr shows the pattern clearly: precision and speed on a specific, well-defined task, not a general-purpose surgical assistant.
Regulatory infrastructure is catching up in parallel. Poland’s regulatory sandbox model under KRiBSI, and similar sandbox approaches emerging elsewhere in the EU, give teams a supervised way to test AI tools before full market authorization, which should gradually reduce the legal uncertainty that 86% of surveyed countries currently cite as a barrier.
Expect continued growth in administrative automation, since it carries a lower regulatory bar than diagnostic tools while still freeing meaningful clinician time. Expect slower, more deliberate growth in diagnostic and surgical AI, gated by the conformity assessments that MDR/IVDR and the AI Act now require together. The health systems that move fastest over the next few years will likely be the ones that treated governance as infrastructure from the start, not as a final compliance step.
An Engineering-First View on Clinical AI Priorities
The technology is rarely the bottleneck anymore. Governance, workflow fit, and training are. Clinical teams that co-design pilots with engineers from day one, rather than handing over a finished tool, end up with systems people actually trust and use.
Regulators, clinicians, and engineers each see a different failure mode coming. Listening to all three early is cheaper than fixing any one of them after launch.
— Vlad
How DevPulse Supports Healthcare AI Projects
Moving from a promising pilot to a system clinicians rely on daily takes engineering that treats compliance and clinical validation as part of the build, not an afterthought. We work with health organizations on exactly that handoff.
- Full-cycle AI development, from discovery through deployment, with EHR interoperability and secure integration built in from the start.
- Lifecycle monitoring and post-market performance tracking designed to meet MDR/IVDR and AI Act expectations together.
- Regulatory-aware engineering practices informed by project experience with healthcare platforms.
Typical engagements move through discovery, a constrained pilot, scale planning, and ongoing support, the same sequence outlined above. If your team is ready to scope a governed pilot, our technology consulting services are a practical place to start that conversation.
Sources
Use these primary sources when preparing a compliance or technical plan rather than relying on secondary summaries.
- News
- New AI technique could make minimally invasive surgeries safer and more precise | MIT News
- Health
- Gov
This article is general information, not a substitute for advice from a qualified doctor. Consult a qualified healthcare professional about your own circumstances before acting on anything here.
FAQ
How can artificial intelligence be used in healthcare?
AI supports clinicians in imaging, automated measurements, administrative documentation, scheduling, and remote monitoring rather than replacing clinical judgment. In practice, tools like the Platforma Usług Inteligentnych support radiologists by automating measurements and protocol adherence.
What is the main application of AI in medicine today?
Imaging and diagnostic support remain the most mature applications, alongside administrative automation for scheduling and documentation. Surgical navigation research, such as the xvr technique from MIT, points toward more patient-specific tools reaching clinical use in the coming years.
Do doctors actually use AI in practice?
Yes, in specific and narrow ways rather than for autonomous decision-making. Physicians in Poland already use AI tools such as the Platforma Usług Inteligentnych for imaging support, and adoption is expanding as regulatory sandboxes reduce legal uncertainty.
Which healthcare jobs does AI threaten to eliminate?
AI is reshaping specific tasks within roles, such as drafting documentation or flagging imaging findings, rather than eliminating clinical professions outright. The bigger workforce challenge identified by the WHO European Region is a lack of formal AI training, which affects how safely clinicians can oversee these tools.
What regulations govern AI tools used in healthcare?
In the EU, high-risk clinical AI must meet both the AI Act and MDR/IVDR simultaneously, with coordinated conformity assessments as described in European Commission guidance. National frameworks, including Poland’s act on AI systems, add oversight bodies like KRiBSI and regulatory sandboxes for testing pilots.
