How is AI being used in healthcare?

We often hear about the potential for AI in healthcare, or how it could transform organisations like the UK’s National Health Service. The UK has set up the NHS AI Lab with areas of focus such as AI imaging, AI ethics and regulation. In the USA, AI in healthcare is expected to save over $200bn from annual medical spending.

How do we expect AI to make a difference in healthcare? We expect to see impacts such as:

• better diagnoses
• personalised support for patients
• faster drug discovery (both in identifying drug targets and analysing preclinical literature using natural language processing)
• more efficient planning of clinical trials with less waste and more informative outcomes.
• greater efficiency across the system
• mobile imaging units that can identify signs of diabetes in retinal scans, bringing first world healthcare to low income countries and villages that are otherwise hard to access

With all these exciting possibilities, we could be forgiven for asking, where is the AI revolution in medicine that we were expecting?

The potential applications of AI in healthcare are backed by evidence. However, uptake and integration into existing healthcare systems has been slow and results have been mediocre.

What is slowing down the adoption of AI in healthcare?

There are a number of factors that make the adoption of AI in healthcare more difficult than in retail.

On the positive side,

• healthcare rightfully has high barriers for the introduction of new methods and technologies. Any AI used for diagnosis or treatment would have to be approved as a medical device by bodies such as the US Food and Drug Administration. There is no such requirement in many other industries.

However, there are a number of more frustrating obstacles.

Inaccessible data

Developing an AI model for healthcare would require a large volume of data. This data exists but is highly fragmented and often inaccessible: governments are aware that their citizens value their medical privacy. Data is often in text format rather than a more easily usable structured format.

Clinical study results, including adverse events such as asthma attacks, strokes, and deaths, are often reported in huge documents for analysis. This data is highly confidential as well as unstructured and in natural language. At Fast Data Science we have undertaken consulting engagements where we have categorised or anonymised clinical reports using NLP.

Fortunately, in the UK we have initiatives such as OpenSAFELY, which makes sensitive healthcare data available to approved research groups within a siloed environment. Advances in NLP allow researchers to handle larger volumes of text data.

Regulation

Regulatory authorities may be slow to approve new innovations in healthcare and may not have the expertise to fully assess the new AI tools. It is also crucial to assess models for AI bias, safety, and transparency. Ideally, countries would work together on regulation of AI in healthcare and create international standards, but we’re not there yet.

Impracticality, and technology not living up to the hype

The idea of an AI replacing radiologists (or any other skilled profession) is still a bit sci-fi. An AI might be able to analyse and classify an image, but can it interact with the patient, or even take the image like a human operator? We are still a long way from skilled medical professionals being replaced by AI.

Conclusion

AI is having a growing impact on healthcare, with the potential to improve diagnoses, personalize treatment, and streamline processes. Here’s a breakdown of how AI is being used and the challenges to wider adoption:

Promising Applications of AI in Healthcare

• Enhanced Diagnostics: AI algorithms can analyze medical images for faster and more accurate diagnoses.
• Personalized Care: AI can help tailor treatment plans to individual patients based on their medical history and other factors.
• Drug Discovery: AI can accelerate drug development by identifying potential targets and analysing research data.
• Clinical Trials: AI can optimise clinical trial design and analysis, leading to more efficient and informative results.
• Improved Efficiency: AI can automate tasks and improve overall healthcare system efficiency.
• Remote Diagnostics: AI-powered mobile units can diagnose diseases in low-resource areas.

Challenges to Wider Adoption

• Stringent Regulations: Healthcare has high safety standards, and AI tools used for diagnosis or treatment need rigorous approval processes.
• Data Fragmentation: Valuable healthcare data is often scattered, siloed, and not readily usable by AI systems due to privacy concerns and unstructured formats.
• Regulation Challenges: Regulatory bodies may struggle to keep pace with AI innovation and lack the expertise to assess AI tools effectively. Additionally, there are concerns about bias, safety, and transparency in AI models. Ideally, international collaboration is needed to establish clear standards for AI in healthcare.

Overall, while AI holds immense promise for revolutionising healthcare, overcoming these challenges is crucial for its successful integration into existing systems. There have also been some positive developments in improving AI in healthcare, such as initiatives to make anonymised healthcare data available for research and advancements in Natural Language Processing (NLP) that can handle large amounts of text data.

References

[1] The AI doctor will see you… eventually, Economist (2024)