Q&A: An AI tool to speed up breast cancer diagnosis
Tobias Rijken, the co-founder and CTO of Kheiron Medical Technologies, speaks to Med-Tech Innovation News about Mia, an AI breast cancer diagnostic tool.
Tell us where the idea for Mia came from?
The idea behind Mia came from the personal experiences Peter and I shared from an early age. We both grew up in a medical environment – Peter’s mother is a radiologist and I have always been surrounded by doctors growing up.
As a result of our upbringing, we witnessed first-hand the daily struggles and challenges radiologists face, and very soon we came to realise the significant difference technology can make to help improve healthcare and patients’ lives for the better.
There are two main issues that Kheiron wants to address: The global shortage of radiologists – particularly breast expert radiologists – and speed up breast cancer diagnoses for the 2 million women that are diagnosed with breast cancer annually.
In the UK alone, one in seven women will develop breast cancer in their lifetime. And while the survival rate is high, breast cancer is still the fourth most common cause of cancer death in the UK.
Research has always shown that prevention is better than cure - with the risk of dying from breast cancer being reduced by 41% over 10 years by participating in regular mammography screenings. It’s within this context that we’ve invested heavily in developing Mia – our Mammography Intelligent Assessment – which is ultimately trying to address both of these issues at once.
Mia can read breast mammogram images to the same level as a radiologist, supporting doctors in making the critical decisions in screening, and improving the experience for women.
It is an AI solution for breast cancer diagnosis – can you tell us the exact nature of the technology behind it?
Mia is a software that helps radiologists detect breast cancer earlier. What we’ve created is software that is specifically designed to analyse standard Full Field Digital mammography (FFDM) images in a breast cancer screening setting, assisting radiologists in detecting breast cancer.
Prior to engineering and training Mia, we had to make sure that we understood the radiologist’s tasks, the data they were working with daily, and their workflow. That was the only way we could identify where AI would be most useful and helpful to radiologists, and ensure it was deployed to be part of their workflow. Essentially, Mia is the result of our combined machine learning methods, data science and radiologists’ expertise.
Like with any technology being deployed in healthcare, Mia has had to undergo robust testing to ensure safe deployment. Our commitment to rigorous clinical testing through large scale data trials is at the heart of what we do and how we develop and scale our technology.
Through robust trialling with large and diverse data sets, we’ve also made sure that Mia was trained to be generalisable, meaning that our AI solution will work to the same level of performance for any woman anywhere no matter the variance in breast density, ethnicity or screening service set up.
How does it make a radiologist’s job easier?
We know there’s a global shortage of radiologists, and Mia’s role will ultimately address this issue by acting as an independent second reader – but always with a human as part of the process.
In the UK, every woman’s screening mammograms are read by two independent radiologists who each decide independently if she needs to be recalled for further examination or not.
This is called double reading screening and is considered the gold standard when it comes to cancer detection and minimising the number of women who are called back unnecessarily. What Mia is essentially trying to do is replace one of these readers, so they can dedicate their time to more valuable tasks.
In the US, we are adding Mia as the second reader. The US has what is called a “single-reader workflow.” But for a while now American radiologists have seen the value of UK’s and the EU’s “gold standard” approach when it comes to breast cancer screening, as it is more likely to detect early, and smaller, breast cancers than a single reader.
We want to make it clear that Mia will not replace the radiologist’s role. Instead, it will support radiologists, reducing 40% of their valuable time where they can dedicate to other complex tasks.
Mia was built to work in tangent with human doctors. And our research trials continue to demonstrate that Mia can act as an independent second reader while ensuring the final decision is made by the human doctor.
Tell us about your footprint in the UK. You recently received a grant to roll out the technology across the NHS?
Kheiron operates in sites across the UK, Europe and the US. The UK has always been a promising market for us because of the government support and incentives it grants to foreign academics, like Peter and I, to develop and commercialise our research.
It’s equally important to add that our footprint in the UK dates back to pre-2016 when Kheiron was founded. As mentioned, we’ve always been focussed on telling the story of Mia as a safe, effective and thoroughly validated AI solution – and at the heart of this is our commitment to robust trialling, which started in 2012.
In November 2020, we presented the first internal results of our large-scale study that’s been carried out between 2012-2019. It included over 40,000 Mammograms from 3 NHS Breast Screening Centres. The results revealed the positive potential of Mia in acting as an independent second reader in breast cancer screening.
This research forms part of our much larger clinical study of Mia that is being led by an Independent Contract Research Organisation and consists of 250K mammograms from 7 sites – in the UK and Hungary. Again, we stress clinical rigour needs to be at the heart of any AI deployment.
We’ve been encouraged by the support we’ve received from both the UK government and the NHSX in recognising our work and funding the rollout of our project across the NHS and the next phase of our trials. In September, we were awarded new funding as part of the UK’s first Artificial Intelligence (AI) in Health and Care Awards, which will help the deployment and clinical studies of Mia across 15 sites in the UK.
It’s a huge vote of confidence in the success of Kheiron and a clear testament to the work, expertise, and commitment our team has made over the years in creating, fine-tuning and scaling Mia.
It seems that some of Mia’s attributes when it comes to image quality and screening can be used in other areas aside from breast cancer. What are your future plans for the product?
Right now, our main priority is to help radiologists detect breast cancer earlier, so our focus is on ensuring Mia is ready and safe for deployment across the NHS and beyond.
The support from the government and the results from our latest trials will help us with the next stage of Mia, which will be its roll-out across NHS sites.
We believe this comes at a pertinent time, with the UK, and the rest of the world, facing an enormous backlog crisis because of the pandemic.
We’ve seen how many women have missed their breast cancer screening during lockdown– with an estimate of 11,000 women living with undiagnosed breast cancer in the UK alone – and we know it will take months for the NHS to resume its services and treatment to pre-covid times.
That’s why now, more than ever, it’s important for us to tell the Mia story – as a safe, effective and thoroughly validated AI solution that can speed up diagnosis, alleviate staff pressures and ensure women everywhere can receive the treatment they need, sooner rather than later.
Anything else you’d like to add?
Anyone working in AI has a special responsibility to make sure that we are inclusive - from the cases we train our models on to the populations we serve. At Kheiron we recently announced a new partnership with Emory University to develop Mia for African American women.
Mia will evaluate data from prior mammograms on over 50,000 African American women who have been screened at Emory Healthcare. Research shows that this population has been historically underrepresented in breast cancer research and in development of artificial intelligence. This new partnership is designed to ensure that Mia minimises bias and is an important step forward for AI to be beneficial for any woman, anywhere, attending breast screening.