Dear Saskia,

re number 3. above:

The obvious barrier to AI in healthcare is data privacy. As long as people see private data being used as a commercial good (cf. social media platform business), this will most likely pertain. The European GDPR is an important step to change this, and thus build trust in the longer run.

Meanwhile, the interesting question is if and how a concept of blockchain and Distributed Ledger Technology could be applied to manage and track permits of individual persons for the use of their private data by third parties in the healthcare sector.

Regards, Norbert Jastroch

Thank You for this interesting question(s), i'll try to answer individually after each question.

1. What is the current healthcare ecosystem? Define different stakeholder groups in the healthcare system.

• What are the key healthcare stakeholder groups that will play a part or be impacted by AI?

The most important word in this question is “key”. 

Key number one is money, since without that there will not be any development with healthcare systems. It is the matter of country’s government (at least in Finland) to decide of the amount of money that can be spend to healthcare during the next fiscal period and for the regional municipal areas that produce the healthcare services in their area with that sum, either via public services or via private services.

When speaking of an AI, one key group should be the system provider that can offer and deliver appropriate artificial intelligent system that can help healthcare provider and customer.

Legislation is also a main key in order to use AI within healthcare. GDPR has usually been seen as an obstacle where data needs to be processed, when in reality it can be seen as the standard how to make things properly. We really need a proper set of ruling and clear legislation with examples in order to trust for handling of information in a such a delicate environment as healthcare.  AI cannot function properly without data, and which data is allowed to be handled within AI system, or where this data can be stored? A big cloud provider in the US may not be the proper place? These are just a scratch of the surface and healthcare personnel as well as patients need these issues solved before to trust in these new AI systems.

• What are the main stakeholder groups in healthcare overall?

   For a Healthcare environment, there are two main stakeholder groups:

          Customers (patient) and healthcare personnel.

For secondary groups, there are owners (private or public), regulators (law), officials (EU, Governmental or Municipal) and relatives of the patient.

2. How is the healthcare ecosystem changing with AI? E.g. how AI deployment (suggestions and predictions of outcomes) will this reshape how healthcare is done. How will AI connect the different stakeholders and what are the main wins (speed, reducing waste) and/or losses. Theoretical assumptions - “hack the dynamic”.

• How will these different stakeholders be impacted by AI?
• What is the visionary potential envisioned through AI?

For a patient, AI system in healthcare may provide better diagnostics, cheaper medication and longer stay at home instead of hospital bed or nursing home. It can also provide better privacy when people are not handing the data. In a worst-case it may also mean loneliness if the only company is an AI system at home.

For a healthcare personnel, AI can function as an extension for their abilities or free their time to more important tasks. For example doctor or nurse can consult AI system for a certain dilemma (for example with the use of medication combination) and use the AI speech user interface for automated logging and documentation. It can make documentation easier, finding information more efficient and get rid of keyboards that are known to be bacterial hives in the hospitals. Easy use of the system will result better outcome once AI system is launched.  In a worst case, personnel can be replaced with AI automated services in order to reduce the costs of the healthcare.

For a provider AI can make a difference via a cost of the healthcare. Costs may also rise if more diseases are recognised on population with the AI systems and they need to be treated accordingly. Nightmare could be a situation, where AI is put to assess the cost of each medication or treatment of individual and making decisions how everyone will be treated. Would elderly people still get as good healthcare as young and more productive citizen? But it is not all bad for a provider, since logistics with AI and governing systems with AI can give significant savings also without reducing anything from the patient of from personnel.

3. What are the main requirements for any AI system in healthcare to function properly? E.g. data privacy etc. Concerning both the development and deployment/maintenance aspect, including mindset shifts that need to happen (not just operational but also cultural and governance).

• What are the main barriers to adoption for AI in healthcare?
• What are the main requirements for ensuring a sustained and optimal use of AI in healthcare?
• What are the main steps that can be taken to overcome these barriers and promote AI adoption in healthcare

The biggest barrier at this moment is probably the lack of knowledge what can be done with the data. Access to the data is vital for all AI systems and now healthcare personnel just don’t know can they handle the data in most efficient way, if at all.  GDPR is all good, but there really is not much interpretation how it can be used in reality. Clear set of thumb rules is needed here.

I’d like to suggest one of Robocoast innovations in here – Living Lab Concept. Bringing up new AI systems in the healthcare is not a simple task, and they should be thoroughly tested before taking them in the wide use. In a Living Lab (for example) a Hospital can open their GDPR compliant data as test environment under local DIH (like Robocoast DIH) for companies that develop AI systems. When developer can test with the real data, their success rate is higher to provide more mature and working systems for hospital and healthcare system in general, who can learn to use new AI system in the same time and give feedback for the developer. This really is a win-win situation for all participants and usually these Labs promote future co-operation as well. Robocoast DIH has also a possibility to educate personnel for the AI and the use of the system since we have two universities in our ecosystem.

Steps in short:

1. Clear and good instructions for the healthcare so that organisations can and dare to use their own data.
2. There really need to be true and real cases that can be modelled and then resolved.
3. Creating Living Lab environments, where latest AI-propositions for systems can be tested, operated, learned and used.

4. What is the market size and ratios and the potential for AI adoption in this space? Demonstrate how important healthcare is as a market for investment (not just manufacturing or automotive).

There’s really not much data available for me to crack this question. The average should be around €5000 per person in a year, and this could be brought down by better diagnostics and healthier population. However not everything should be measured by business side or by money. The health can be priceless and – as seen on some country surveys – gives better change to be happy in life.

Anyhow , healthcare in general is probably one of those fields that will benefit most of the use of an AI, and as a market it is one of the most interesting ones.

wbr.  Kai Salmela , AI Specialist  Robocoast R&D  DIH

Hi Saskia,

Is there a deadline for feedback?

Thank you

3. What are the main requirements for any AI system in healthcare to function properly? 

Startups and corporations need to clearly define areas for AI products within their business, what value could be created and how hard would it be to capture it.

Based on that, they should place the AI products within the correct place in the customer/patient journey and capture lowest hanging fruits first. This way, the would build up a success stories and it will open doors for futher AI adoption.

Right now, corporations try to cooperate with startups with low risk, delegating too little resources and attention to those products, which can result in poor pilot's KPI.

Another improtant aspect is to prove the quality of the AI product (and clinical safety) on a day-to-day basis. Typically, AI products tend to evolve quickly, and the quality is assessed only once before the pilot. The quality of the recommendations should be monitor on a regular basis. It will increase the understanding and adoption of the AI products in healthcare. 

Clients want to see the reasons why model/product showed particular recommendation -  the models need to be interpretable to a user without an AI background. If the user doesn't understand the recommendation - he/she will never us it. Building an understanding of the products is a must for AI product adoption. 

One of the objectives in the classification of stakeholders is analysis of their needs and provide the best solution possible or to initiate related research to find the best solutions possible. In general, Healthcare Stakeholders Groups that can be effectively supported by AI can be divided into two groups:

• A. Subjects and groups of subjects
• B. Institutions and organizations

These general groups can be further divided into:

A. Subjects and groups of subjects:

• patients (from different perspectives: in clinics, in a hospice, at home, mobile, etc.),
• disabled users (from different perspectives: with cognitive problems, with motor problems, etc.),
• older adults (from different perspectives: living independently at home, living in senior houses, etc.),
• relatives and friends of subjects from the above groups (including informal caregivers)
• healthy subjects (from different perspectives: health prevention, management of long life in health, etc.),
• healthcare professionals and related personnel (including physicians, nurses, paramedics, physiotherapists, etc.),
• other groups (age-related groups, job-related groups, etc.).

B. Institutions and organizations

• hospitals and clinics
• hospices
• social care institutions and organizations,
• rehabilitation centres
• governments and local-government organizations
• insurance companies and organizations (including governmental agencies)
• other (founding organizations, etc.).

This simplified classification is of course not complete. However, I would like to underline that there are important stakeholders in healthcare that are not only patients, doctors or hospitals. 

It is also important to underline that the health services have been recently provided more often in various environments, not only in hospitals or clinics. Due to the technological progress (including AI and Robotics) more diagnostic, therapeutic (including rehabilitation) and preventive activities will be provided at homes, at work places, in smart environments, also using smart wearables or implants. Today, automated external defibrillators (AEDs) are standard equipment in places of our work, in public places, etc. Why not having surgical robots in the future work places or homes (e.g. to sew a wound)?

Finally, it is interesting to ask about another future aspect: should we consider AI stakeholders? Today we are sometimes discussing about an “AI doctor” (vs. a human doctor), etc. What would be the needs of the “AI doctor” to provide the best health service possible?

It would be also interesting to prepare an analysis of healthcare stakeholders from the AI perspective similar as prepared from the AAL perspective (Ambient/Active Assisted Living), presented, for example, in this document: http://www.aal-europe.eu/wp-content/uploads/2015/02/AALA_Knowledge-Base…

Dear Colleagues,

I find this discussion very interesting.

• I agree with Jacek‘s list to identify stakeholders. One could add pharmaceutical companies, universities, researchers, education and training, producers of instruments and materials etc.
• Additionally, I suggest to look at the processes among the “stakeholders”, how they act and react, information and knowledge being produced, recorded, passed on, uncertainties, dealt with, decisions made etc., and flows of money among them. Some sort of Health 4.0.
• Regarding the barriers, yes, privacy of data is one issue. But – whenever someone says, data is the “fuel” of AI, I react: this is a limited view of AI (following the current dominance of deep learning). AI is also based on knowledge: facts, rules, common sense, experts’ knowledge, physical knowledge, physiological knowledge, mathematical knowledge (including differential equations – do you know how to solve those?) and so on…, AI is based on knowledge acquisition, modelling, and processing, and how to handle uncertainties, and so on.
• Yes, we can start doing business with “deep learning” (e.g., classifying x-ray pictures – but this is not “understanding” x-ray pictures!!). Hence the real barrier is: the complexity of health.
• So there is a lot to be done – and for many (if not the most) applications in healthcare we dream of, we still need a lot of interdisciplinary research, collaboration. That is the core issue.
• In such early phase, when technology is not yet mature, unclear how it will fit the processes in healthcare, it is very difficult to assess “markets”, since the organisation and processes between stakeholders may change, the same way as IT is changing the structure in companies.
• Healthcare will improve massively when the connecting IT landscape is well integrated and updated regarding cybersecurity (hospitals are suffering from inhomogeneous networks which even include old Win85 PCs). This does not require AI. Building good databases is also not yet AI. And, as I have heard, hospitals may save 5-10% in materials when their logistics are modernised. So there is a lot of potential – even without AI.

Best,

Uwe