How does your previous experience in medical imaging relate to Lucida? 

For 10 years I worked in clinical research for a company managing imaging in clinical trials, joining as one of the first employees and growing the company to nearly 50 people and £6m turnover. We worked with biotechs and pharma on clinical trials, providing technology and training to transfer medical images and provide expert quality assurance and imaging end points using a combination of machine learning and expert radiologists.

This breadth of experience has enabled me to talk to radiologists about imaging, specifics of MRI scanners, diffusion weighted imaging and so on.  I understand the language they use; my background in physics also helps on the MRI side!

This experience has proven particularly useful. I know many of the clinicians that Lucida is working with, and developed software under the regulations required for medical devices.  This added to my previous background in software and machine learning in other regulated industries from aerospace to banking. Of course, it has been a steep learning curve, however, I feel I’ve hit the ground running. 

Lucida Medical has just been awarded CE mark approval. What is your vision for the company over the next 2-5 years?

The CE mark is a particularly important milestone because it means we can market Prostate Intelligence in the EU and UK as a medical device that can be used in clinical practice.

Over the coming years, we will establish our product as a device that clinicians routinely use. We have plans to extend its capabilities, particularly around how radiologists interact with it. This will be an ongoing task.

Whilst prostate cancer is important, it is not the only cancer. We are already working to extend our platform to other cancers that are amenable to analysis by MRI, and other applications across oncology, and in due course we will expand geographically.

With regards to the patient journey, there is the initial diagnosis of the cancer, followed by staging, and finally the observation of response to treatment. With our automation of the image analysis process, we can add a lot of value to each of these steps. If we can speed things up, we can offer better clinical pathways for patients.

Which is the one AI breakthrough you will be on the lookout for in the upcoming year?

We will see incremental improvements in algorithms, will continue to see the so-called deep learning models develop further. At some point we will see a step change in the way we approach artificial intelligence and machine learning.

The current paradigm is that we know what the answer is because the human expert has told us, or we have measured the outcome retrospectively.

People working on dealing with less data. With medical imaging you need thousands of examples, sometimes this data is simply not available to us. People are working on synthesising more data out of existing data, or training something with less data.

Those are ongoing studies. Somebody somewhere will be doing something incredibly exciting, studying a PhD and thinking these convolutional neural networks are not the best way to approach this problem. They will have a moment of inspiration end eventually we will all say ‘did we really used to do it like that’ – in some years’ time of course.

The momentum behind Machine learning is so big, no company above a certain size, and that’s a relatively small size, isn’t currently looking at their data and wondering how to use machine learning to better understand our customers to target them more effectively, improve our products, get better feedback etc.

One might expect convergence between machine learning and environmental considerations. How we use machine learning to limit our impact on climate and carbon neutrality.

Do you think AI will replace radiologists?

Geoffrey Hinton, no relation, often considered to be the father of machine learning and artificial neural networks, famously said in 2016 that “we should stop training radiologist now because in five years’ time they won’t be any need for them”, perhaps somewhat tongue-in-cheek. Anyone who is working in machine learning knows what it can and cannot do. Machine learning will find its niches in clinical pathways to speed the process up. It can replace some skilled but repetitive tasks, but it cannot yet (and nowhere near yet) replace the human in critical thinking, problem-solving and creativity.

What machine learning can do for radiologists, is alleviate them of some of the tedious stuff in the same way that basic computer technology has impacted office life. Machine learning can help enormously with productivity and decision support, but the radiologist is still the person that fully understands the anatomy and can deal with things that are out of the ordinary, like unusual cases and incidental findings.

Clearly AI won’t replace radiologists, but our vision is to use it to make them more accurate and enable them to focus their skills and expertise on the clinical decisions that will most benefit patients.