Researchers at Khalifa University in Abu Dhabi have developed iGenRARE, an agentic AI platform intended to support doctors in diagnosing more than 7,000 rare diseases.
The technology analyses information collected from a patient’s hospital visits and compares it with evidence from medical literature and specialist rare-disease databases.
It then provides doctors with a ranked list of potential diagnoses, explains the evidence supporting each suggestion and recommends further tests to confirm or rule out a condition.
Dr Aamna Al Shehhi, assistant professor in biomedical engineering and biotechnology at Khalifa University, is leading the project and says the system was designed to address one of the biggest challenges facing patients with rare diseases: the lengthy search for an accurate diagnosis.
“Usually, one of the major problems with rare diseases is late diagnosis,” she told The National. “For some rare-disease patients, the journey to the correct diagnosis can take five years or more.
“For a progressive disease, those years matter enormously – the condition may worsen, irreversible complications may develop and precious opportunities for early treatment can be lost.”
During those years, patients and their families may move from one doctor or specialist to another, undergo numerous tests and investigations, only to end up with no clear explanation of their symptoms. “What is the probability that a doctor has seen a similar case before? It will be rare,” Dr Al Shehhi said.
Agentic AI uses several specialised systems, known as agents, which work together to complete a complex task. In iGenRARE, one agent can assess symptoms while others examine clinical notes, genetic variants, laboratory results, medical images and published scientific evidence. A central reasoning system then connects the findings.
“Our system goes beyond genetic information,” Dr Al Shehhi said. “We incorporate the different hospital visits, treatment, the progression of the disease, imaging and other types of information in one platform.”
Ranking based on probability
The first stage screens existing medical records for signs that a patient could have a rare disease. If a case is flagged, the second stage conducts more detailed analysis using the additional clinical information and scientific literature.
Rather than presenting one condition as a definitive diagnosis, iGenRARE ranks several possibilities according to the evidence and its level of confidence.
It can then tell the doctor which additional investigations may be needed to distinguish between diseases with similar symptoms.
“The system will provide a ranking based on probability,” Dr Al Shehhi said. “If it needs to be more confident about a specific disease, it can ask for additional tests. It can tell the doctor to perform a particular test to confirm whether it is that condition.”
She stressed that the platform is intended to assist clinicians rather than replace their judgment.
The system has been retrospectively evaluated using US medical data covering 20,238 patients, including 5,067 patients with rare diseases and a control group without rare diseases.
Using information from only a patient’s first hospital visit, the model achieved 68 per cent accuracy in distinguishing rare-disease cases from controls, Dr Al Shehhi said.
That increased to 88 per cent after analysing information from the first two visits. When the patient’s entire hospital history was assessed, accuracy reached 91 per cent.
The results do not yet represent performance in everyday hospital care.
The researchers now want to conduct a silent clinical evaluation in which iGenRare runs in the background on real clinical cases, without influencing patient care. This allows them to compare iGenRare's recommendations with clinicians’ decisions, and evaluate its accuracy and safety before moving towards active clinical use.
“Even though the model has done an amazing job on the US data, when we put an AI system into real-life practice, there might be a difference,” Dr Al Shehhi said.
“We need a silent clinical trial, with the doctor using it in parallel with their practice, to see how the tool performs. If it does not perform as expected, we will understand why and improve the AI model.”
A hospital version of iGenRARE would be installed within the healthcare provider’s own secure infrastructure rather than be made freely available online, Dr Al Shehhi said.
This would allow the platform to analyse sensitive patient information while keeping it within the hospital’s systems.
A separate public version could eventually be developed using only publicly available information, although that is not part of the project’s present stage, she said.
Emirati Genome Programme
Dr Al Shehhi said the platform could also complement projects such as the Emirati Genome Programme by interpreting genetic variants alongside symptoms, disease progression, tests and imaging.
The researchers have also started exploring how AI could help identify treatments for rare diseases after a diagnosis has been made.
Developing a new medicine for a condition affecting only a handful of patients is often commercially difficult. The next phase of iGenRARE will therefore examine whether existing medicines could be repurposed to treat rare conditions.
The longer-term ambition is to help doctors develop personalised treatment plans reflecting the circumstances of each patient.
“After diagnosis, we need to try to find a treatment,” Dr Al Shehhi said. “Some rare diseases already have effective treatments, while for many others, treatment options remain limited or unavailable.
“We are now expanding iGenRare towards drug repurposing -identifying existing medicines that could potentially be used for rare diseases – and, in the future, towards supporting more personalised treatment recommendations, because every patient is different.”
