At the Huawei Connect 2018 conference, held in Shanghai last week, one of the many tangible applications of the use of AI that was explored was how the technology  can be used in the healthcare sector.

Admittedly, the use cases across the sector are vast, all with a broad focus on improving patients’ health. However, there are also highly specific ways to do that.

Cancer in the crosshairs

In an interview with Dr Sabrina Reimers-Kipping, the cofounder of Fuse-AI explained that the company’s artificial intelligence software solution is finely focused to specifically help doctors diagnose prostate cancer.

“In the field of radiology, there is a large need to come up with new solutions that enable radiologists to cope with their growing workload,” she explained. This is due to the fact that the number of available radiologists is not keeping pace with the need for scans for prostate cancer.  Complicating matters is the need to ensure that diagnostic quality doesn’t suffer because of overloaded radiologists.

To this end, the software, which employs machine learning and deep learning, creates markings on MRI scans and labels structures that are likely cancerous, a process that can be time consuming for human beings.

“Human inspection of data is what takes long, usually 25 minutes to check the images, with additional time needed to create the markings that identify where there are anomalies on the scan. We hope that by using our software to do the labelling and generate a report, this process will save doctors time,” she elaborated. Reimers-Kipping explained that the algorithm used by the software to achieve this serves as a second opinion to the doctor, which takes minutes to be generated, and thus can be a valuable aid in the diagnostic process.

How it works

The software relies on some 2000 MRI images sourced from Radprax in Germany.

Reimers-Kipping stressed that it wasn’t just important to have large amount of imagery data, but that those images are of high quality, pointing out that a neural network or AI system can only be as good as the data it is fed.

‘Teaching’ the software what it needs to attain higher degrees of accuracy is a work in progress. Reimers-Kipping elaborated that the lower limit is 91% accuracy, which is currently what a radiologist can accomplish. However, she stressed that the idea isn’t to replaced evaluations done by a person, but rather to complement it with AI and machine-learning, so as to achieve higher levels of accuracy in less time.

This is especially pertinent when it comes to certain cases where the start of cancer can be spread on the outskirts of the prostate, and are more difficult to detect. As well, doctors could utilise the software in those cases where they know that human analysis on its own would be likely to be less accurate.  

AI, together with the Cloud

Echoing Huawei chairman Eric Xu’s call for there to be greater synergy between different technologies in his top ten changes that need to be made to enable AI, the AI software also works with the cloud.

For Fuse-AI, it is Open Telekom Cloud, located in Germany, that the company sends the encrypted MRI scans to, where they are analysed by the AI algorithm. It is there that the metadata and markings are added and then sent on to the doctor.

The reason for choosing Open Telekom was simple - complying with the regulatory requirements for handling health related data in the EU made sense, particularly as the company looks to expand its solution into other European countries.

Even as Fuse-AI’s technology is focused on a particular type of cancer, Reimers-Kipping noted that there is plenty of scope for similar AI-based solutions to identify a variety of other types of abnormalities and different kinds of cancers. As for the future of Fuse-AI’s solution, she enthused that the company does envision offering it on a global scale in due course.