AI in Healthcare: Challenges and Solutions

According to the statistics, US medical experts were not very enthusiastic about AI implementation in the healthcare industry last year. Only 23% of respondents believed AI technologies would be more advantageous than harmful. 42% thought AI in healthcare would be both beneficial and damaging. That means, despite the AI hype in all spheres of life, people still feel concerned and doubtful about some issues. In this article, experts from the Belitsoft Software Development Company summarize the challenges of AI in healthcare and share insights on how to overcome them.

Input data quality

Generative AI uses data to produce content of any kind. Tools that generate medical diagnoses like Atopic dermatitis app, read X-ray images, and MRI scans need impeccable data to avoid faulty results. AI-powered tools use information from electronic health records (EHR) and electronic medical records (EMR) systems. More sophisticated tools can generate personalized diagnoses with the help of genetic, environmental, and social data.

For instance, risk scoring for diabetes contains a questionnaire about age, waist circumference, genetics, blood glucose, and diet. Machine learning algorithms may further use this data to predict the risk of developing diabetes in various population groups.

Medical staff are pressed for time, so the quality of data they upload into EHRs often leaves much to be desired. Deceptive abbreviations, irrelevant terms, and inaccuracies lead to misunderstandings among medical providers.

Automation tools may help with these issues by complying with standard data requirements and checking the relevance of the information. As the Chief Innovation Officer with 19+ years of expertise in the HealthTech domain, I constantly emphasize the necessity of implementing business intelligence (BI) tools and data analytics for medical providers. EHR systems with embedded analytical tools can process large volumes of data, extract the necessary information, and provide it to experts for decision-making. Thus, customized EHRs with BI tools help medical organizations in the following ways:

●     Well-maintained medical data becomes a source for further implementation of AI technologies.

●     Doctors decrease paperwork and use their time more effectively.

●     Marketing and management teams analyze reviews and communication channels and build a strategy to cut costs and scale.

Disclosing personal data

AI technologies are visible first and foremost in chatbots on websites. They release the staff from answering routine questions and organize the workflow by transferring specialized requests to appropriate experts and departments or scheduling appointments with doctors. However, people do not trust chatbots, as they are afraid of data leakages and phishing. As a result, patients are not likely to share sensitive data online.

To deal with that issue, businesses should guarantee safe assistance for their customers. AI-powered tools should comply with regulations like the upcoming EU AI Act.

Secondly, users of the chatbots on medical sites should feel secure that their requests are not used by machine learning algorithms for training purposes if they are against that. For instance, ChatGPT will soon supplement Apple’s Siri, and Apple has promised to protect the privacy of its users. Medical sites should allow its visitors to turn the AI model training off if they wish to safeguard their data.

Biased conclusions

A lack of information AI tools can learn from may lead to irrelevant results and risky consequences. For example, there is a lack of information regarding African Americans in the US EHRs. The reason lies in an insufficient number of patients who participate in clinical trials. Besides, some population groups live below the poverty line or reside in remote areas. As a result, AI algorithms learn from the data, which mostly contains facts about white, well-off US citizens.

Health inequities are among the main issues the US healthcare system has to address. According to Deloitte experts, AI has to become an integrated part of medical services, leading to equitable health care. Thus, AI algorithms should regularly be updated and tested with a focus on the following:

●     Data strategy. Gathering data should happen across all geographical locations and with an unbiased approach.

●     Improved testing. Testing machine learning algorithms is different from traditional testing principles. AI generates new content based on updated data in its ‘brain’. Therefore, the expected results are different every time.

●     Careful monitoring. Medical data can quickly become obsolete. It is essential to track the changes and react to updates timely.

Staff competence

Dealing with AI tools requires certain skills and, therefore, might lead to possible risks. It might be challenging and confusing for medical professionals to perform their duties with AI assistance. Certain practices and training are required. Secondly, when making a diagnosis with AI-generated systems, doctors have to supervise the results and double-check conclusions to avoid serious mistakes in the treatment plans. It adds pressure on doctors, as they have to be responsible for the machine as well.

Combating this issue is a long process. Including AI technologies in medical education is the first step on the way to introducing those tools to future doctors. For instance, the students of the NYU Grossman School of Medicine have the following opportunities to collaborate with AI on their lessons:

●     Students can ask the AI to provide them with educational resources, e.g., infographics, videos, clinical results, medical literature, etc., regarding a particular patient.

●     Students interact with virtual patients and get feedback from the AI. It evaluates their performance as well as the empathy demonstrated towards a patient.

●     AI software can control students’ performance, offer additional training material, and aid with goal accomplishment. Consequently, it assists human coaches and trains young doctors to interact with AI tools.

Final thoughts

AI tools have the potential to drive medical progress forward. They can facilitate issues such as staff burnout, administrative task overload, and long waiting periods for medical care.

However, there are still risks and challenges that experts and medical providers have to tackle. Dealing with the issues mentioned in the article requires a multidisciplinary approach and the joint efforts of medical experts and development teams. AI implementation in the medical domain, first, needs a well-thought-of roadmap that will include best practices in software development and data science. The second part of the strategy should comprise mentoring and staff training to ensure motivation and low stress levels.

"Dmitry Baraishuk is a partner and Chief Innovation Officer at the software development company Belitsoft (a Noventiq company) with 19+ years of expertise in digital healthcare, custom e-learning software development, and Business Intelligence (BI) implementation'