On 16 March 2022, Deeper Insights founder and CEO Jack Hampson delivered the webinar Transforming healthcare using AI - How to move from ideation to a more scalable AI solution alongside guest speaker Darren Wilson from Smith & Nephew. This article is a summary of the topics covered.

If you work in healthcare, pharma or life sciences and are looking to get an AI project off the ground or scaled successfully, continue reading.

The AI landscape

We're seeing more companies with AI in their roadmaps than ever before, in part due to increased access to open-source technologies and off-the-shelf solutions. As AI technology becomes more obtainable and affordable, a more democratised relationship is evolving.

We are delighted to be witnessing an increasing familiarity with AI and consequently an increase in trust, confidence and understanding of the value. The market is maturing from a speculative to a much more solutions-led approach. This focus on solving specific problems generates much greater value.

The value of AI for every stage of the healthcare journey is becoming more evident. In recent years, headlines heralding AI's contribution to drug discovery have become more commonplace. Additionally, AI is becoming more integral to healthcare in the clinical setting, improving the service of patient aftercare and predicting health outcomes.

As the role of AI across healthcare increases, so do the rules and regulations around it. The increasing interest and involvement of the FDA and other regulators helps build trust around AI's use in clinical settings - an important, progressive step for the industry.

Big breakthroughs in healthcare AI

In 2021 alone, AI technology M&A deals worth over £32.5bn were announced globally, up 31.5% year-on-year. (Source: Gartner)

Industry leaders are making visionary leaps in pushing the boundaries of what is possible with AI, from drug discovery to personalised medicine and aftercare support.

Learn more about some of the big breakthroughs:

• DeepMind - AlphaFold: Using AI for scientific discovery
• Alzheimer’s drug designed using Exscientia’s AI technology enters clinic
• Healx’s Rare Treatment Accelerator 2.0 opens to unlock the power of repurposed drugs for rare diseases
• Nuritas - AI-powered peptides primed to shake up nutrition

The AI climate for healthcare

Cloud computing has been a significant enabler for AI adoption and innovation, changing the way we host data and train models, and ultimately increasing the affordability and accessibility of AI for healthcare.

There have been huge breakthroughs in Deep Learning, Transfer Learning and Active Learning in recent years with a knock-on effect for managing unstructured data. We are now able to use datasets previously viewed as unusable (for example, videos and images) in a way that was previously unimaginable.

Thanks to Transfer Learning we can use smaller healthcare datasets, as well as adopting various methods to augment or synthesise data to generate larger data sets from smaller ones and use this information to train a model with greater accuracy and confidence.

With these advances, the healthcare industry has made great breakthroughs and the climate is prime to nurture further advancements in the coming years.

Getting started with AI in healthcare

To get over the starting line there are a number of areas to consider:

1. Data sets - You need fit-for-purpose, compliant data sets to get started. There are privacy laws around the use of patient data so you must adhere to the rules and regulations surrounding this.
2. End-users - Gather buy-in and educate your end-users, whether that’s patients or physicians. It can be very advantageous to gain feedback from early adopters also.
3. Subject matter experts - Secure input from experts in AI and/or machine learning to accelerate the process and ensure direction and best practice.
4. Business Value - Get buy-in from all the stakeholders and ensure an appropriate budget is in place. Think about all the experiments and various clinical trials and the costs associated with these.
5. Mindset - Making sure that everyone that will be involved across the project is on-board and can visualise the impact benefits.

Avoiding the pitfalls in AI projects

Bear in mind the following to avoid a failure:

• Purpose - Begin with purpose, and ensure that you design a project to be specific to meet demonstrated user needs, there needs to be a use case beyond the appetite to play with AI. What is the problem we are trying to solve and what does AI bring to the solution?
• Metrics - Ensure you have metrics in place for quality assurance and measure the performance of AI early on in projects. When dealing with a new project, it’s difficult to know how much data is required. Augmentation techniques allow you to create more data and train the model, but ultimately, until you begin to train the model you cannot accurately predict how much data is required. So early performance assessment will allow you to improve incrementally. Think about how you will stress test the data?
• Technology - Utilising new tools will bring with it unforeseen challenges, as a new structure will change existing dynamics. Think about how you will collect the necessary data and ensure that you create a sustainable system. If you are collecting data from specific hardware (x-ray, RGB scans etc…) then think about how you can get the highest quality version of that image to ensure success.
• Communication - Incorporating AI into healthcare requires the collaboration of many teams, you’ll have the hardware team, data scientists, engineers, project managers, business leaders, practitioners and patients etc. Communication breakdown is an easy yet avoidable route to project failure.

Scaling considerations

While AI is advancing rapidly there are not that many applications that have come to market as yet, but scaling is an important consideration from the offset.

Scaling considerations need to bear in mind IT infrastructure to deliver the AI pipeline in a working environment, which of course brings with it many challenges when concerning integration with hospital systems, and the associated data privacy and security concerns and constraints.

For AI-influenced surgery, for example, to determine the true reliability of the machine algorithms, stress testing needs to progress from cadaver environments to live surgery and ensure that practitioners are comfortable. Data can be collected fairly autonomously during a regular procedure.

Iterative input from all stakeholders and end-usersis also crucial and making sure performance decisions are captured at every stage of the process, backed by trials and evidence.

Ongoing costs are a critical consideration to ensure you have a sustainable long-term solution in place.

Conclusion

To conclude, moving an AI project from ideation to a scalable solution takes a lot of planning.

We are reliant on data, and working against a checklist for data quality control and meeting regulations.

We need good communication channels across all stakeholders and be ready to adapt if the models are not training satisfactorily, if more data is required, or even if large scale changes are required to the entire project.

It takes a lot of hard work and persistence to take AI through the various gates of clinical trials and evidence the performance of the AI solution.

However, we believe the transformative power of artificial intelligence can solve the toughest human problems, and we’re excited to be part of it.