The Clinical AI Pathway Guide is a practical and structured guide designed to support healthcare professionals, researchers, and companies in the development and implementation of AI solutions. It offers a clear, stage-based framework to help healthcare teams move from early ideas to real-world clinical use. The model consists of six stages, each outlining key actions and identifying relevant stakeholders essential to AI development.
“This pathway is very good, simple and easy to follow and will help many companies when planning these first time!” Juuso Juhila, Director of Clinical Operations at Aiforia
Although the guide is structured in stages, AI development is rarely linear. This iterative process model visualizes the real need for moving back and forth between stages—allowing for continuous learning, refinement, and better alignment with clinical, technical, and regulatory demands, even after deployment.
Structured Clarity: Follow a clear, structured guide from concept to clinical deployment.
Maturity Tracking: Use the TRLs in the guide to evaluate and communicate the maturity of your AI solution.
Stakeholder Alignment: The guide helps you to identify who should be involved and when, fostering collaboration and accountability.
Clinical Relevance: Through the guide you will engange with tasks that will help you ensure, that your AI development stays grounded in real-world healthcare needs and practices.
Risk Reduction: The guide will help you to address ethical, technical, and regulatory factors early—avoiding surprises later.
What are “TRLs”?
A central feature of the guide is the use of Technology Readiness Levels (TRLs) to assess AI solution maturity, track progress, and ensure readiness at each step. The TRLs measurement system ranges from TRL 1 (lowest), where the basic principles are observed, to TRL 9 (highest), where the technology is proven in a operational environment and integrated in clinical practice. EURAXESS: Why using TRLs?
Watch this video to get the most out of the Clinical AI Pathway Guide
The video provides a clear overview of the guide’s framework and practical applications, helping you understand how to navigate each stage and use the guide iteratively.
When you get an idea for a new AI healthcare solution, it is essential to begin with a clear vision and well-defined strategy to ensure the solution is both relevant and effective. This includes defining the clinical need, gathering stakeholder insights, and getting an overview of ethical, regulatory, and practical requirements to guide the project. Early planning focused on market positioning, funding, and team expertise ensures a structured approach, paving the way for a solution that is impactful, compliant, and aligned with real healthcare needs.
Outcome
The outcome will be a well-scoped AI project rooted in a real clinical need, shaped through stakeholder input and close collaboration between clinical, technical, and domain experts. Key ethical, legal, regulatory, and market factors will be identified early, creating a strong and strategic foundation for development.
Conceptualization & Proof-of-Principle Development
Introduction
During Conceptualization & Proof-of-Principle Development the focus is on refining the AI solution’s design to meet clinical requirements and validating its feasibility. This involves defining user needs and technical specifications, creating an algorithm development plan, and deciding on key model parameters like accuracy, usability, and training type. Ethical and regulatory preparations continue, with applications for data access, ethical approvals, and preliminary intellectual property assessments. Early planning of data handling, anonymization of data, and quality standards further support compliant development. Strategic partnerships, an earning model, and regulatory alignment are also emphasized to ensure both functional and regulatory readiness as the solution takes shape.
Outcome
The outcome will be the necessary approvals secured to move forward, including data access and ethical clearances. With support from Senior Experts where needed, an Intellectual Property Rights (IPR) strategy and regulatory plan will be established, ensuring legal compliance and minimizing risks, setting the stage for the next phase of development.
Development from Proof-of-Concept to Prototype
Introduction
When transforming a conceptual AI model into a functional prototype, essential activities include gathering and preparing data, refining the algorithm, and selecting validation methods to ensure clinical accuracy. Emphasis is placed on regulatory compliance (e.g., General Data Protection Regulation (GDPR), EU AI Act) and interdisciplinary collaboration. Additionally, clinical and cost-effectiveness analyses validate the model’s applicability and impact in real-world settings.
Outcome
The outcome will be a functional prototype approved by the your team, ready for retrospective validation using historical data. Preparations for prospective clinical testing will be in place, enabling real-time evaluation. These steps help verify performance, ensure clinical relevance, and mitigate risks ahead of implementation. This process ensures the prototype’s readiness for safe and effective use, setting the stage for live testing.
Validation of Technology Demonstrated in Operational Environment
Introduction
Validating the AI solution in clinical environments is done to ensure the prototype meets the clinical needs. This includes preparing for implementation by engaging users, refining market strategies, and collecting new data to verify performance. Clinical validation and regulatory tasks confirm that the solution meets safety, effectiveness, and regulatory standards. Documentation and transparency are prioritized to support a smooth path to regulatory approval, ensuring the solution is ready for deployment.
Outcome
The outcome will be a clinically validated AI solution tested in real-world settings to confirm safety, effectiveness, and alignment with clinical needs. A functional prototype/Minimum Viable Product (MVP) with a user-ready interface will be developed to gather critical feedback for refinement. Validation data and documentation will support regulatory submissions, including Medical Device Regulation (MDR) compliance, while commercial and technical implementation plans are refined. These actions position the solution for final approval and readiness for launch and deployment.
Product launch & System qualification
Introduction
While preparing your AI solution for market entry is crucial to plan how to launch your product. Key activities include completing the conformity assessment and securing CE-marking. You will also register the product in European database on Medical Devices (EUDAMED) and, if exporting, apply for a Free Sales Certificate.
A business plan will be developed to highlight the product’s clinical benefits to end-users and key opinion leaders (KOL), alongside conducting a Health Technology Assessment (HTA). Finally, planning for implementation ensures that the end-users are prepared, and potential challenges are addressed, paving the way for a successful launch.
Outcome
The outcome will be a fully certified AI solution ready for deployment, having completed CE-marking, conformity assessment, and regulatory registration. A clear business and implementation plan will support uptake, while a Health Technology Assessment (HTA) results and promotional materials will help communicate value to users and decision-makers. The solution is now positioned for integration into clinical workflows and operational use.
Deployment
Introduction
During Deployment, the priority is to seamlessly integrate your AI solution into the hospital’s IT-infrastructure, ensuring compatibility with existing systems, technologies, and clinical equipment, while effectively managing operational challenges. Change management is essential, requiring continuous evaluation and fostering user ownership to embed the solution into daily routines. Collecting user feedback will help improve product quality and maintain data traceability. Post-implementation evaluations are crucial for ongoing enhancement, while post-market surveillance ensures compliance with regulations and monitors product performance. Vigilance measures must also be in place to address safety concerns and uphold patient safety.
Outcome
The outcome will be the successful integration of the AI solution into clinical workflows, supported by continuous performance monitoring and user feedback. Real-world data will be collected to assess clinical effectiveness, usability, and safety. Post-market surveillance and vigilance processes will ensure regulatory compliance and enable ongoing improvements, safeguarding patient outcomes and supporting long-term adoption. Post-market surveillance should be an ongoing process after the AI solution has been deployed as illustrated in The Iterative Proces Model.
Start defining the clinical problem you aim to solve with your AI solution.
Start defining the expected value creation of your AI solution: Will it save time for the clinicians/end-users? Is it cost-effective?
Start defining the current state-of-art to be used as baseline for comparison at a later stage.
Start defining relevant clinical data needed for algorithm development: training and validation.
Have you involved clinicians in the process?
The data set should represent the targeted population
Initiate dialogue with stakeholders or interested parties and AI competent developers/technical expertise in early phase.
Make a stakeholder analysis.
Who are the different stakeholders? Who is important to include?
Examples could be: clinicians, patients, data engineers, UX designers, anthropologists, researches, IT architects and security experts, legal experts, management.
Consider doing a SWOT-analysis on the current state-of-art.
What are the Strenghts, Weaknesses, Opportunities and Threats?
Start collecting feedback from other clinicians/ end-users. Do they also see the problem? If yes, do the find it an applicable solution to solve the problem?
Note: It can be challenging to create understanding amongst clinicians about the impact of an AI solution in an otherwise well-documented clinical field, which can make it difficult for a clinician to evaluate.
Start defining the expected benefits of the new AI-solution compared with traditional Standard Operating Procedures (SOP).
Ensure they are measurable for evaluation after implementation.
Identify the competencies and knowledge that can strengthen your project.
It is important to build a diverse team with both technical experts and clinicians (like doctors, nurses, psychiatrists) who can connect the technical side with the real-world needs of the project, especially for AI in healthcare. A diverse team enhances the ability to progress through various TRLs adapting to evolving needs througout the development process.
Are there internal or external individuals or organizations necessary to ensure interdisciplinary teamwork?
Some of the typical competencies needed in AI-projects include: clinicians, patients, data engineers, UX designers, anthropologists, researches, IT architects and security experts, legal experts, management.
Define Project Owner, Project Lead and eventually describe a Governance Structure for the project.
Integrating a solution into a Hospital’s IT-environment can be a real challenge. To address this, working with Hospital's IT-Departments and companies should be a priority early in the development phase to identify compliance requirements.
Define the roles and areas of responsibilities within the team to ensure alignment within the team.
Define intended use of your solution to qualify/classify your product both as concrete type of medical device and AI system.
Who is using it, who is the target population and what is it used for?
Based on the intended use of the device, your device might be qualified as a medical device under Medical Device Regulation (MDR) or as an in-vitro diagnostic medical device under In Vitro Diagnostics Regulation (IVDR).
Familiarize yourself with the Medical Device Regulation(MDR) / In Vitro Diagnostics Regulation (IVDR) as well as EU AI act definitions, obligations and requirements, as central CE-marking regulations relevant to your product, by using the source document or other guidelines (for example, various ISO standards).
Get the overview of applicable legislation for your product. Due to the variety of product features AI solutions in a medical context could be subject to different sets of laws.
Note: Compliance with the EU AI act, Medical Device Regulation (MDR)/In Vitro Diagnostics Regulation (IVDR) might not be enough.
Familiarize with regulatory strategies in collaboration with a legal expert.
Consider inviting Senior Experts, research, Business Support Agencies as mentors or Advisory Boards to provide input and feedback on the strategy and/or required adjustments.
Start deciding on required level of accuracy/verification plan.
Patient data (hospital's data lakes, biobanks, primary care, social care)
Available commercial data
Open-source data.
Assess the data for any potential technical quality issues.
Note: approval processes can be slow. Contact your data provider early to obtain information regarding required documents, procedures, and expected timeline. You can also get support from the Hospital's IT-deparment and Data Engineers.
Get acquainted with the legislation related to use of data for Research & Development (R&D), EU level and national/ regional legislation.
Clarify if patient consents are needed when data is collected or used for research and development.
At this stage you should consider the requirements associated with a product and its development, as well as the general medical device obligations of manufacturers. Your goal is to develop technical documentation which allows you to pass conformity assessment procedure successfully.
Make a risk classification analysis according to Medical Device Regulation (MDR) / In Vitro Diagnostics Regulation (IVDR) and the EU AI Act.
Make a GDPR risk assessment.
At this stage, have you considered the following?
Identified General Safety and Performance Requirements (GSPRs) related to your product?
Appointed Person Responsible for Regulatory Compliances (PPRC) within your institution?
Registered yourself as manufacturer of medical devices in European database on Medical Devices (EUDAMED)?
Started to implement/adjust your Quality Management System (QMS)? Put emphasis on implementation of risk management and usability engineering concepts into your development work discipline.
According to your GDPR risk assessment identify whether your data needs to be anonymized (data anonymization), pseudonymized (data pseudonymization) or other.
Note: If in doubt, contact legal experts to be sure the data is handled correctly.
Local validation is a critical step in AI algorithm development to ensure context-specific accuracy and reduce the risks associated with unvalidated commercial products. Therefore, it is important to:
Choose your validation methods
Retrospectively?
Statistics?
Choose your software architecture.
Cloud vs. self-hosted?
Using different types of data can help ensuring the robustness of the algorithm
Note: Specificity and sensitivity is a clinical decision.
Evaluation by experts and comparison with diagnoses determined using Gold-standard methods can help determine whether the AI solution reliably detects what was intended and specified in the requirements analysis.
Clearly demonstrate the value of your AI solution. Highlight time savings for clinicians and cost-effectiveness, as these are critical factors in a clinic’s decision to adopt your solution.
Consider doing a Cost-Effectiveness Analysis: Do the benefits provided outweigh the costs of implementation/administration of the new test?
What is the business/economic motivation for the implementation of your new tool? Does the tool improve patient outcome, reduce costs, shorten/reduce hospitalization, improve selection of treatment, decreases morbidity and mortality? Is this achieved without compromising patient outcomes?
Revisit your definition of the intended use of the model under "The problem, the need and the idea" in the "Idea" phase to reevaluate the potential of your AI solution as a new clinical care pathway.
Consider where the new AI solution fits best:
New steps in Testing chain?
Improves current steps?
Replaces current steps?
Support from experts like Health Economic Specialist.
Obtained regulatory expertise as described in the "Idea" phase.
Familiarized yourself with the Medical Device Regulation (MDR)/In Vitro Diagnostics Regulation (IVDR), EU AI Act definitions, obligations and requirements, central CE-marking regulations as described in the "Idea" phase.
Developed technical documentation which allows you to pass conformity assessment procedure successfully as described in the "concept" phase.
From here, you need to:
Start developing performance characteristics related to your medical device. These attributes may include accurary, precision, sensitivity, specificity, durability, and usability.
Ensure your medical device is safe and performing for its intended use.
The most important dimensions are related with performance evaluation (scientific validity, analytical performance and clinical performance).
Document all the activities justifying your compliance with identified General Safety and Performance Requirements (GSPRs).
You want to seek out an Medical Device Regulation (MDR) experts for company-specific challenges on this topic.
At this point it is beneficial to revisit the project plan you made in the "Idea" phase to make an estimate of the potential market, number of potential users, feedback from potential users and more thorough analysis of the competitive landscape.
Assess competition. Has other companies developed a similar solution? What does the customers and other value drivers think?
Conduct design verification of your medical device
Stakeholders:
CompanyRegulatory Expert
Conduct design verification of your medical device
Start taking care of transparency and traceability of your device by complying requirements related with European database on Medical Devices (EUDAMED).
Demonstrate compliance with applicable GSPRs (General Safety and Performance Requirements), especially in term of analytical and clinical performance information (as an analysis of performance evaluation/clinical evaluation) and stability studies (claimed shelf life, in use stability and shipping stability studies).
Once you have ensured transparency and traceability of your AI soultion in line with the requirements mentioned in the "Validate" phase, the AI solution needs to be registred in the European database on medical devices (EUDAMED) to support regulatory transparency.
Identify the first users of the AI solution, define their implementation roles, and establish a plan for managing potential issues during launch.
For concrete steps, see: Guideline for Adoption and Implementation based on Change Management.
There should be explicit guidelines for user training and support, along with steps to assess interoperability with existing hospital systems including training programs for end-users. This would help ensure smooth adoption of the AI solution.
Integrating a technical solution into a hospital’s IT-environment can be a real challenge. To address this, you have been working with IT-Departments and vendors since the "Develop" phase and have identified compliance requirements.
Some laboratory instrument providers are more open for allowing access to their Application Programming Interface (API) or even source code enabling seamless technical integration.
Reach out to the Hospital's IT-Department to ensure a proper technical integration.
To accomplish assessment of change management in hospitals, there should be both continuous evaluation during the implementation process and after each stage, and there should be a sense of ownership, and the new products should be used in daily routines. Therefore, it is necessary to:
Encourage a sense of ownership among staff can be achieved by involving them in the decision-making and planning processes
Integrate the new product into daily routines through adequate training and support can promote adoption and minimize resistance
Receive regular feedback from users should also be gathered to refine processes and enhance overall effectiveness
Note: Since AI tools change, it is important to have a process on how to monitor and collect feedback from the clinicians/end-users (e.g. on malfunction or usability problems) to make necessary optimizations to local environments (which might cause MDR reevaluation) and in the end improve the product quality.
Post-implementation evaluation becomes a key part of the learning process when working with AI solutions and is important for ensuring that the AI solution is adapted and improved over time. Therefore, the evaluation plan that was created pre-implementation needs to be conducted.
Contact Hospital's IT-Department.
Revisit the clinical evaluation plan in the "Develop" phase.
