Using computers to make more accurate diagnoses

Fast Facts

• Title: Automated classification of brain tumours using artificial intelligence
• Lead Researchers: Dr Phedias Diamandis
• Where: University Health Network, Toronto, Canada
• When: January 2019- December 2020
• Cost: £114,750 over a 2 year period
• Research type: Adult and Paediatric, High and Low Grade Tumours, Academic, 
• Grant round: Expanding theories

---

What is the purpose of the study?

In even the recent past, a brain tumour diagnosis was made solely by a pathologist looking down a microscope at a small slice of the tumour on a glass slide. As we know, new information in both technology and science are always expanding our capabilities. In 2016 the World Health Organisation (WHO) released their new recommendations on how to diagnose brain tumours. In addition to descriptions of how the different tumours look down the microscope, for the first time they included molecular tests for some types of tumour. This shows the value of new technologies in making more accurate diagnoses. With this in mind, Dr Diamandis and his team will be developing a complex artificial intelligence (AI) program to provide a step change towards the next generation of diagnostic tools.

Teaching the AI

This form of AI is called deep convolutional neural networks (CNN), and is used to find patterns in images. The first aim of the research is to teach the CNN to recognise brain tumours. To do this the team will use more than 10 million images of different brain tumours. These tumours have associated notes detailing the patient’s diagnosis, survival and therapy response. The AI will then pool all this knowledge to make diagnostic predictions on a new set of images (from prospective patients).

Letting the AI teach us

The team will then use the AI’s machine learning to find other patterns in the images, and associated data, that are too small or infrequent for clinicians to recognise alone. These new insights could be used to make more accurate diagnoses, better prognostic predictions and suggest the best treatment for each tumour.

Ultimately, Dr Diamandis would like to create an automated, cloud-based program so that the AI is available to everyone to use.

Why is it important?

As skilled as they are, neuropathologists are ‘only human’. This means their assessment of a tumour through a microscope is subject to human error. It also means that different pathologists could give different diagnoses for the same tumour. This project opens the door to a new way of making a diagnosis, by using artificial intelligence.

The more we know about a tumour at diagnosis, the better chance we have of being able to target it, with appropriate treatments, faster.

By making the tool widely available, Dr Diamandis would ensure that all clinicians, even those in the most remote settings, have equal access to the very best knowledge to enable them to provide the best care possible.

Who will it help?

This research has the potential to help everyone who has a potential brain tumour. It could mean a more accurate and reliable diagnosis, with better indication of prognosis, and even faster delivery of the best treatments.

While histopathology plays an important role in optimising care for the majority of brain tumour patients, it is long overdue for innovations that yield more objective, timely, and personalised diagnostic information.

Achievements

The Toronto team have developed a highly flexible and nimble brain tumour classifier that is proving capable of diagnosing the specific type of cancer in an automated fashion. Because the specific type of cancer dictates the type of treatment someone will receive, such a tool would ensure each and every person gets the appropriate treatment for their specific cancer. By automating the diagnostic process using state-of-the-art computer algorithms such as artificial intelligence, the tool can help provide the best standard of care and the optimal diagnosis to geographical areas that may not have a pathologist with a brain tumour specialty.

To facilitate the use of the tool in clinical trials, Dr Diamandis has also developed a web-portal that allows the tool to be used by any center in the world. This will be particularly helpful for people in clinical trials as the tool may be able to predict which patients are best suited for a specific clinical trial. The team is now testing the diagnosis tool in another cancer centre 10 hours drive from Toronto. This cloud-based solution will help ensure the clinical community can use the tool for their patients, even in remote cancer centers.
To ensure the product reaches the clinic fast, the team are working with industry partners to help develop the tool into a tangible medical device that we hope will eventually become standard of care.