noun_Email_707352 noun_917542_cc Map point Play Untitled Retweet Group 3 Fill 1

AI research project aims to speed up diagnosis of rare diseases

Teams from Tietoevry Care and Helsinki University Hospital (HUS) set out to tackle a diagnostic challenge with the help of Artificial Intelligence.

Niina Siipola

Head of AI and Data Solutions

Joint research by Tietoevry Care and Helsinki University Hospital is using artificial intelligence to accelerate the diagnosis of certain rare diseases. The project holds promise for relieving patient suffering and reducing one of healthcare’s biggest costs.

Some rare and serious diseases are notoriously difficult to diagnose. Patients who present with symptoms may be repeatedly misdiagnosed as having a condition far more benign than the underlying reality.

When the symptoms repeat – or evolve – the patient may be sent from unit to unit within a hospital and tested many times before the correct diagnosis is reached. For some rare diseases, the diagnostic process can take a decade or more. This not only means years of pain and discomfort for the patient, it’s also expensive for the healthcare system. Rare disease treatment is one of the costliest areas in healthcare.

Teams from Tietoevry Care and Helsinki University Hospital (HUS) set out to tackle this diagnostic challenge with the help of Artificial Intelligence.

In a Business Finland funded research project – named eCare for Me – the joint team has spent the past couple of years developing algorithms and data-lake capabilities to enable quicker and more accurate diagnosis of three groups of rare diseases: glomerulonephritides (kidney diseases), myositides (muscle diseases) and vasculitides (diseases of the veins and arteries).

Big data in a secure pool

The team selected these three disease groups – all of which are inflammatory conditions – as they are the most difficult and expensive for the HUS system to diagnose and treat. Each year HUS diagnoses approximately 540 new cases of glomerulonephritides, 60 of myositides, and 170 of vasculitides.

“We obtained a research permit to access the HUS data pool of about 3.5 million patients, from which we identified the data of 60,000 rare-disease patients as relevant to the study,” says project lead Niina Siipola, Head of AI Solutions and Data Services at Tietoevry Care.

“When you have such a large amount of lab data available, you can train algorithms to select certain data points and from there to predict whether a patient has a specific rare disease. The aim is to be able to diagnose these patients much quicker than is currently possible, which will lead to better health outcomes and help to lower the cost of care.”

“During 2022 we got the first version of the model trained and it yielded good results. Now we’re retraining it to become even better,” says Siipola.

One of the core principles of the project is to conduct the research and build the environment in compliance with both GDPR and Finnish regulations on the secondary use of health data. There’s vast opportunity to improve diagnostics and accelerate medical research through the use of patient data, but the work needs to be done in a modern environment with the right security protocols.

Enabling quicker diagnosis

“These three disease areas are among the most devastating that exist. They cause considerable pain and suffering for the patient. Vasculitis is the one we attacked first, as it’s so complicated to diagnose and treat,” says Mikko Seppänen, Head of the Rare Diseases Unit at HUS.

Vasculitides diseases present as systemic inflammation of the arteries or veins. This inflammation can occur anywhere in the body, potentially causing life-threating conditions such as thrombosis or infarction (the death of tissue due to a lack of oxygen).

While the exact cause of most vasculitides remains a mystery, the majority of cases can be attributed to a combination of genetic predisposition and environmental factors. The conditions often seem to be triggered by viral or bacterial infections.

“Nowadays our medical practices are so siloed that it can be many years before specialists reach the right diagnosis,” says Seppänen. “For example, when only the inner ear and the sinuses are inflamed, an Ear, Nose and Throat Specialist will not necessarily think of vasculitis.”

“The data show that patients with vasculitides progressively increase their use of medical services for about 10 years prior to diagnosis. The longer it takes to diagnose any of these diseases, the more destructive they become.” he says. “Patients in all these disease groups end up being treated in intensive care if diagnosis becomes extremely delayed.”

The research project is continuing through 2023, with the team further training the model and developing features that will allow doctors to receive notifications. The team’s vision is that the AI capability will eventually be able to proactively alert doctors to the possibility that one of their patient’s may be suffering from a rare disease. This would mean integrating the algorithms with doctors’ patient notes for pattern recognition.

“We’re still training the model and doing a lot of studies related to natural language processing so that the AI can find certain indications in doctors’ notes,” says Siipola.There’s a lot of work to be done, but preliminary findings show the algorithms getting close to our goals.”


Read more about our sustainability work and how we help to create purposeful technology that reinvents the world for good.

This case is published in our annual report 2022 – read the whole report here.

Share on Facebook Tweet Share on LinkedIn