Hospitals Are Already Digital. The Next Step Is Artificial Intelligence
The past digitalisation of the healthcare system have paved the way for artificial intelligence. While AI has the potential to streamline processes, logistics, and free up human resources for patient care, its implementation must be well-considered so as not to disrupt current processes.
“Microscopes did not evolve for over 200 years. Then, with the introduction of digital scanners at the turn of the millennium, researchers and clinicians were able to scan a whole piece of tissue at once.”
This is how Richard Lindelöf – who is the senior vice president of clinical product management at Visiopharm – explains the great paradigm shift in pathology. Ultimately, the change in technology and digital imagery led to improved workflow and created new possibilities for data storage and information management.
“When tissue samples are sent physically, it costs money, takes time, and sometimes the samples disappear. With digital technology, we can scan samples in one location and look at them from the other side of the globe”, Lindehöf explains.
While digitising the sample-images provided benefits from the start, it has also paved the way for the next big paradigm shift in the health sector: artificial intelligence.
“It is a prerequisite for image analysis that the samples are digital. Fortunately, since the advent of digital scanners, we have received much more data, which means that today we can provide AI tools to pathologists who can both save them time and make them more proficient,” Lindehöf says, who is Senior Vice President President of Clinical Product Management in Visiopharm today.
The computer becomes the doctor’s autopilot
With the right digital data – and enough of it – deep learning techniques can be used to train artificial intelligence to recognise diseases from an image of a tissue sample. Compared to human researchers, AI can be used to analyse an enormous number of images in a matter of minutes.
Illustrating the point, Martin Kristensson, who is the company’s senior vice president of sales for the European region, explains:
“If you imagine that the sample image represents the entire globe, the relevant point that doctors need to find to diagnose a patient is the size of Denmark. Although doctors are capable of doing that well, with image analysis and AI we can find the relevant point faster and send it to the doctor, who then makes a decision based on it. We are not trying to replace pathologists, but to streamline, augment their work, and improve the accuracy of diagnoses.”
He compares the AI tool to the autopilot in an aircraft: If you get into an airplane and are told that the autopilot is not working, but the pilot is here, you stay on the plane anyway. But if the pilot is sick, you do not trust the autopilot alone.
In the same way, patients should not be diagnosed by Visiopharm software alone, but it should help the physician make the diagnosis.
“We are not trying to replace pathologists, but Augment them,” Kristensson says.
And in fact, it seems that the combination of pathologists and computers not only frees up time, but also improves the number of correct diagnoses.
“We see cases where pathologists agree with each other on the diagnosis 60% of the time. But if we do image analysis and afterwards get pathologists to look at the images, they agree in 90% of the cases,” Lindelöf says.
A rightness of fit
By using AI and deep learning, Visiopharm provides the most comprehensive solution for medical diagnoses based on image analysis. Currently, it is widely used to diagnose breast and lung cancer, but there is nothing to prevent it from being used for other types of cancers and tissue samples.
Of course, it is one thing to train artificial intelligence and validate the results, but it is quite another to ensure the AI tools fit into the everyday life and flow of a hospital.
“It’s quite complex and we see it as our biggest task. We don’t just understand the technique – we also understand the customer’s needs so we can find the right solution. And fortunately, we are a very young company that dares to think out of the box, so doctors can have new opportunities,” Lindelöf says.
Visiopharm faced significant challenges introducing its technology to hospitals in 2013. Despite the fact that the company already had its advanced image analysis software running with researchers, it had to adapt the solution to clinical settings.
Reflecting on this, Kristenssson explains:
“Although our technology is among the most advanced in the world, you can’t simply expect that it will transfer seamlessly into any environment. It must be built into a workflow that is useful in everyday life. Even the most advanced AI needs to be packaged into a lightweight user interface that just works.”
By bringing all the hospital stakeholders in Denmark together, Visiopharm succeeded in finding the optimal solution to this issue.
Even so, not all hospitals are making full use of the solution because infrastructure still needs to be put in place. However, it is already showing good results – including “fewer malpractices and misdiagnoses,” according to Kristensson.
Training the missing pathologists
Visiopharm’s solutions are already widely used by pathologists in the United Kingdom, Belgium, Switzerland, Portugal, Sweden, Norway, Finland, and Denmark, and more products are under development. Kristensson notes:
“The application is broad: Breast cancer, lymphatic, skin cancer – anything can be analysed with the right amount of data. It basically takes 15 minutes to create a new functional algorithm. But to optimise, verify, and validate it to ensure that no mistakes are being made, that can easily take a year. Implementation adds another year.”
As new AI is validated and introduced into clinical settings, pathologists can free up their time for what they do best. And that’s good news, given the global shortage of expertise in this area.
“The great future of the technique is that we can make AI do the processing, but also give a digital second opinion afterwards. Increasingly, this will enable pathologists to catch up on the huge backlog that they lack the capacity to cover today,” Lindelöf concludes.