Barriers to Integration

Although AI is driving important advancements in healthcare, there are still barriers that are slowing its incorporation into the industry.

“Integrating AI into healthcare systems comes with several challenges, such as the need for standardization of data and the development of algorithms that are explainable and interpretable,” Brockett. “To overcome these challenges, it is essential to have collaboration between healthcare providers, researchers, and policymakers to develop guidelines and regulations that ensure the safe and effective use of AI in healthcare.”

Firstly, there are important ethical issues to consider when aiming to achieve the full potential of AI in healthcare.

This includes addressing the use of patient data, safety and transparency, algorithmic fairness and biases, and data privacy.

Privacy regulations can make it difficult to collect and pool healthcare data. With especially strong privacy concerns in health care, it could be too difficult to use real, high-quality health data to train AI models as quickly or effectively as in other industries.

“Privacy of personal health information is a tremendous concern and one that should never be taken lightly. In that same vein, there can be a tendency for machine learning algorithms to be biased based on the training information they receive. Therefore, there are ethical concerns around the under-representation of certain patient populations in training data. These concerns extend to the entire healthcare field, not just AI,” Brockett said.

Additionally, the regulatory approval process for new medical technologies takes some time, and the technology is reviewed meticulously. Innovations can take years to undergo the approval process.

Healthcare providers may also hesitate to adopt new technology for fear of law implications so that liability concerns may be a barrier.

In terms of technical limitations, AI algorithms require a lot of computing power, storage capacity, and specialized hardware and software, which can be a challenge for healthcare organizations with limited IT resources and infrastructure.

Furthermore, healthcare data is often put into different systems and formats, potentially hindering the deployment of AI algorithms that rely on large and diverse datasets. 

“To overcome these obstacles, healthcare providers must have confidence in the AI tools they use and ensure that they provide accurate and reliable information,” Kim said. “AI tools must be compatible with existing healthcare systems and workflows, which can be a complex and time-consuming process. At Stanford AIMI, we address these challenges by developing, validating, and disseminating AI tools based on rigorous science.”

Moreover, the actual adoption and integration of AI is difficult when it requires significant organizational and cultural changes, including training and education of healthcare professionals and changes in clinical workflows and decision-making processes. 

Of course, people on the receiving end of healthcare services must be comfortable with the use of AI as well. This presents a further string of challenges.

According to Pew Research Center, about 60% of Americans would be uncomfortable with a provider relying on AI in their own health care.

People are typically resistant to change, especially when it comes to healthcare. When new technology is presented, it can create hesitations. Thus, patient reluctance is a major barrier to implementing AI in healthcare.

Generally, the real-world use of AI is a challenge in itself. For example, almost none of the ML tools recently developed to tackle COVID-19-related problems had a very significant impact.

Overall, while AI has the potential to revolutionize healthcare, its integration presents several challenges that need to be addressed to ensure its safe and effective use in patient care.