The families of patients with rare genetic diseases have, historically, struggled for years – or even for entire lifetimes – to get a correct diagnosis for their disease. But these often gruelling “diagnostic odysseys” might soon be a thing of the past, now that researchers are using artificial intelligence (AI) to accurately detect rare diseases through facial feature recognition.
A global research team which includes Shahida Moosa, Associate Professor of Medical Genetics in the Division of Molecular Biology and Human Genetics at Stellenbosch University, has come up with a new tool that can help clinicians and researchers diagnose a rare disease far more quickly than in the past, based on an AI-driven analysis of facial features.
The tool, known as GestaltMatcher, is featured in a paper which was recently published in the prestigious Nature Genetics journal. The tool was developed by a team of researchers from the University of Bonn, in Germany, and other institutions, with Moosa’s team being the only group contributing from Africa.
Moosa, who is second author of the paper and whose student, Kimberly Coetzer, is a co-author of the paper, said the GestaltMatcher is an exciting new tool.
“It is going to change the way we practice medical genetics, as well as how we do research,” she said in an interview.
Moosa, also a senior consultant in medical genetics at Tygerberg Hospital, directs the new research programme on Rare Disease Genomics in South Africa. Through this programme, which falls under the Faculty of Medicine and Health Sciences’ new Biomedical Research Institute, she aims to change the lack of focus on rare diseases, particularly in Africa, where many patients pass away without ever hearing a name for their condition because rare diseases are “complex” and “under-studied” in Africa.
Interviewed about her role in the paper on GestaltMatcher, Moosa said that most rare diseases are genetic in origin and many of those have associated distinctive facial features, or a so-called facial gestalt.
“It is often with the facial features where we start in trying to find a diagnosis but it is not always easy.
“For instance, we have seen hundreds of thousands of people with Down Syndrome, so it is not difficult to make a spot diagnosis – even in a shopping centre or in the streets. However, there are many much rarer diseases, even so-called ultra-rare diseases, which even an experienced dysmorphologist may never have encountered in his or her clinic. It is for this reason that we are increasingly relying on help in the form of AI.
“GestaltMatcher is what we call next generation phenotyping, which means that the AI is not only able to ‘look’ at clinical features on the face and come up with a differential diagnosis for the patient’s condition, but it is able to ‘match’ the facial gestalt of two unrelated undiagnosed patients. The ability to make these matches based solely on the facial gestalt is unique.”
Moosa said a big benefit of the GestaltMatcher tool is that “it doesn’t rely on being trained on dozens of patients in order to make a diagnosis or a match. It is able to see a patient’s features once and then match it to the very next patient it sees with similar facial features. That is beyond what the human experts can do.”
She said most syndromes which present with unique facial features would be amenable to diagnosis using GestaltMatcher. “We already have web-based and app-based tools, which help us in the clinic every day, but GestaltMatcher takes this to the next level, because, without much training, it can match those who are still undiagnosed. For example where a syndrome hasn’t been seen before or is undiagnosed, GestaltMatcher can match two patients with similar features. A match in this scenario points us towards a common diagnosis for these patients. Novel syndromes can be delineated using GestaltMatcher.
“It can also group patients according to how closely their facial features resemble those of identified syndromes. This helps us to look at genomic sequencing data and to see if the patient for whom we seek a diagnosis has a variant in the same pathway as the genes responsible for the known syndrome, thereby cutting down the search space for the underlying genes considerably.”
Moosa said the new tool is going to “change the way we practice medical genetics in a big way.
“With the growing number of syndromes and ultra-rare syndromes, it is impossible for human experts to be an expert in every one of those syndromes, so we will rely on tools like GestaltMatcher in the future to be able to get to a diagnosis earlier.
“GestaltMatcher is one of the pieces in the puzzle to drive the mission to get a diagnosis as soon as possible. It is also a tool we will use for novel syndrome discovery and novel gene discovery. This is especially exciting for us in Africa, as our patients have not been sufficiently represented in other tools.
“This is one of the tools we will increasingly use to shorten the long, sometimes painful, odyssey that patients go on to get a diagnosis for their rare diseases.”
Moosa, who has been involved with the team of researchers for some time, said she is pleased the paper is finally out.
“The best thing about the GestaltMatcher is that it’s not behind a paywall. We’ve given all the data to a non-profit organisation (Association for Genomic Diagnostics) and made it available for clinicians and researchers to use to drive the field forward.
“I’m so excited about that aspect because, using the tool and the data collected, opens up a whole new avenue for us to be able to help our patients in the clinic and for data scientists and researchers to use it and make better tools of it.”
Read the original article here.