Efficiency of AI has been identified in eye diseases ⁷ , breast cancer ^{8, 9} , and skin cancer ¹⁰ diagnosis when using data mining, natural language processing, and machine vision methods. Furthermore, glucose level prediction is one of the use cases where AI methods have been utilized successfully ¹¹ . AI methods can be used for tuberculosis diagnostics ¹² and prediction of psychosis ¹³ . AIDI platform can implement these studied AI methods to provide direct access to utilize these methods for customers and end users.

Personalised medicine is tailoring of medical treatment to the individual characteristics of each patient. The approach relies on the understanding of how a person's unique molecular and genetic profile makes them susceptible to certain diseases. Personalised medicine also focuses to provide detection of the diseases at an earlier stage when treatment activities can be applied more effectively ¹⁴ . AIDI can monitor patient diagnosis and data and based on data provided to pre-trained AI methods AIDI can suggest most effective treatments and/or medication to patient.

AI methods to detect medication errors and increase medication adherence are used in healthcare sector. AI can be used to create screening system to detect potential medication errors and generate alerts to clinical decision support (CDS) systems and electronic medical record (EMR) systems ¹⁵ . Machine vision and neural networks have been used efficiently to monitor medication intake by patients to enhance medication adherence. AIDI can provide access to these methods as a service to be utilized by healthcare service providers.

In clinical trials it is essential to identify potentially eligible patients to create more comprehensive clinical trials with lower costs. AI can be used to help in clinical trial design and to find patterns in large datasets for identifying most suitable patients before clinical trials start and AI can be used also for enabling more effective and accurate monitoring of patients during clinical trials. By using AI methods increase in enrolment in clinical trials has been observed. Furthermore, utilization of AI methods increases identification of eligible patients for clinical trials ¹⁶ .

On-going 2020 Covid-19 pandemic has increased the need to predict the progression of epidemics and pandemics. AI methods can be used for different epidemics by combining and analysing information of spatial spreading and behaviour data of the epidemic. As a result, estimation and prediction of epidemic spreading can be achieved ¹⁷ . Real-time forecasting of epidemic outbreak with AI have been used in COVID-19 forecasting in China with high accuracy ¹⁸ . Furthermore, AI methods can be used in managing on-going epidemic outbreaks by providing methods for identification of on-going epidemic outbreak, diagnosing and prognosing infections, and identifying possible therapeutic options ¹⁹ . AIDI methods can provide end users the solutions which utilize data from public and private health databases and data lakes to predict possibility for epidemics.

AI has been used in connection with medical imaging for a long time and the use of AI methods has been extensively studied and proven to surpass human diagnosis in practical clinical work. In the field of medical imaging the AI technologies, such as DNN and DL, can produce remarkable improvement in healthcare outcome and they have proven to provide enhancement in speed, accuracy, and cost reduction in interpretation of medical images ²⁰ . Medical image diagnostics can be utilized in healthcare for multiple cases, such as for identifying cardiovascular abnormalities, detecting common eye diseases by optical coherence tomography, detecting musculoskeletal injuries, identifying neurological diseases, identifying thoracic complications, and in oncology for screening common cancers ^21‐23 . AIDI can provide these validated methods to end users for instant image analysis. Our plan is to provide AI methods for medical image diagnostics as among the first of AIDI services.

AI utilization in health monitoring and preventive health consist of solutions for continuous health status monitoring, healthcare assessment tools, and symptom checking solutions which can be used by patients and healthcare professionals periodically or continuously. AI is also commonly used in personal health monitoring and giving personalized suggestions for preventive health based on user parameters, such as activity tracking, nutrition, sleep analysis, weight, body composition, and vital signs, such as blood pressure and blood sugar. Personal health apps are the most common health monitoring solutions which collect data from user smartwatch and smartphone and utilize AI to provide analysis and suggestions for end users. There are also clinical remote monitoring solutions which utilize AI, such as cardiology monitoring, sleep monitoring, and posture monitoring ^{24, 25} . AIDI can provide periodical/intermittent analytics from the collected data and provide analysis to healthcare services development companies or research organizations.

Healthcare services utilize AI based virtual nursing assistants in various use cases. By using virtual nursing assistants, hospitals will be able to reduce sudden hospital visits and reduce the load of healthcare professionals. Virtual assistant applications can listen, talk, and give advice/recommendations based on patient conversation, health history and health data of the patient. Common virtual nursing assistant solutions provide pre-diagnosis for patients and healthcare professionals based on the healthcare assessment questionnaire and patient history data before entering the primary care ^{26, 27} .

AIDI can be used to validate usability of new AI methods. AIDI will contain a growing number of AI methods for healthcare use cases. AIDI also provides access to training data and actual data. New AI methods can be evaluated and validated by comparing new AI method outcome (i.e., identification of accuracy, specificity, or sensitivity) and compare results with previous AI methods results. With this evaluation and validation service third-party AI method developers can create better AI methods which can be utilized in AIDI platform by companies or research organizations. Evaluation can be done by a specific multidisciplinary team and by scoring system in AIDI. This scoring provides end users valuable information about usability of selected AI method in specific use case.

As an example of AIDI we present the process flow of use case where clinicians want to analyse radiology image to identify possible lesion from the image. Steps are explained and shown in Figure 3: 1.

User sends radiology image to AIDI with information about use case (example case breast cancer analysis).

Advantages for actors in healthcare

When utilizing AIDI in healthcare use cases we can provide immediate benefits for all actors in care process and service development. For software and services development, there is less time used for training AI methods, since AIDI developers can use ready-made AI methods and modify methods and algorithms to meet their own use case specific needs. There is also no need to search for training data for AI algorithms, because AIDI contains interface for training data repositories. For healthcare service providers and caregivers, there is instant prediction, analysis, or pre-diagnosis possibility with the software which utilize AIDI platform features. AIDI can provide enhancement in diagnosis accuracy and diagnosis speed. For patients, healthcare customers, and people who are tracking their wellbeing, AIDI provides accurate and instant diagnosis, prognosis, and predictions based on personal health data and verified and trained AI analytics from large datasets.

The savings potential when using AI in healthcare is very high. In USA annual savings are expected to be 150 billion USD by 2026 ²⁸ . When comparing this savings potential to total US healthcare expenditure in 2018 (3600 billion USD with 4,6% annual growth) we expect that US healthcare expenditure will be 5150 billion USD in 2026 ²⁹ . With these values we evaluate that AI utilization can bring 2,9% reduction in healthcare expenditure in the US. We evaluate that same cost reduction percentage can be achieved also globally.

AIDI benefits can be presented by the situation where state-of-the-art AI methods are integrated into AIDI and these services are utilized in healthcare in full scale. Benefits can be introduced in values, such as improvement in care outcome, cost reduction in healthcare, quality-of-life improvement among patients, and reduction in workload among healthcare professionals ³⁰ .

For evaluation of AIDI methods in healthcare we also suggest a first version of the weighted scoring system for AI benchmarking. Our novel scoring provides benefit assessment for evaluating the implementation of new AI method in certain healthcare use case. Output is informed as AIscore. Same tool with modified parameter information can be used to evaluate AI method benefits in other business sector use cases. Parameters of the equation are: Scientifically Proven Improvement compared to traditional method (SPI), Training Data Score (TDS), Cost Reduction (CR), Quality-of-Life improvement (QoL), and Workload Reduction (WR). Parameters and weights will be inserted by evaluation team. Evaluation team has researchers and professionals in a particular field of healthcare. Parameter values inform parameter impact, effect, positive outcome, or improvement in certain use case of AI method. For example, parameter SPI can be set by specificity and sensitivity of AI method. Furthermore, parameter TDS can be set by evaluation training data quality and amount of training data parameters for the specific healthcare use case. Parameters CR, QoL and WR will be scored by evaluation team. Each of the parameters will be using weighted coefficient (W _n) based on the importance of parameter for specific use case. There is also acceptance parameter (AP) expressing whether method can be used for specific use case or not. AIscore will be developed and evaluated in future AIDI research projects.

By using AIDI and its validation of AI methods by AIscore, we can provide centralized AI distribution and utilization platform where all actors can benefit from effective and evaluated AI services. AIDI platform can also contain sales channel for evaluated AI methods which can be easily adopted for healthcare industry purposes. AI method developers can provide methods through AIDI as “Google Play” style where AI will be openly available.

Concerns of AIDI adoption in healthcare

Although AI methods in AIDI can provide scientifically proven enhancement in diagnosis and predictions accuracy, we must keep in mind that AIDI will only provide information and guidance for healthcare decision makers. AIDI platform or AI methods overall are not yet allowed to provide direct diagnosis. Healthcare professionals will always be responsible for providing diagnosis and final decision. There are also challenges which can emerge when AIDI services are utilized in healthcare. These issues are market acceptance and end user acceptance. Possible reasons for these issues are arising from the lack of curated healthcare data. Moreover, previous studies show that there might be resistance from healthcare professionals and patients to trust AI based decisions and diagnosis. Furthermore, people think that there is less “human touch” in care process when AI takes a greater role in healthcare. Moreover, shortage of regulations in AI in healthcare and data privacy can cause challenges. All these issues and challenges should be considered in planning phase of the AIDI project to enable acceptance of AIDI services to all end user groups and to all regulatory authorities.