John was 68 when subtle changes in his behaviour began to raise concern. Initially, these signs were easy to dismiss – misplacing everyday items, forgetting recent conversations, or struggling to recall familiar names. Over time, these incidents became more frequent and harder to ignore.

Despite growing concern, the journey to diagnosis was far from straightforward. Like many families, John and his daughter faced delays, uncertainty, and multiple referrals before receiving a formal diagnosis of Alzheimer’s disease, a common cause of dementia that affects memory, thinking, and behaviour. It took three years to finally reach a diagnosis.

John’s story is not an uncommon one. One of the biggest barriers to a diagnosis of Alzheimer’s disease is that it can look very different for different people, in some cases this is reflective of the different causes of the disease.

“In our memory clinic, we see a wide spectrum of symptoms among patients with Alzheimer’s disease” says Professor Jason Warren, Consultant Neurologist at the National Hospital for Neurology and Neurosurgery and Professor of Neurology at the Queen Square Institute of Neurology. “While many people experience classic signs like short-term memory loss and confusion, others present with more subtle changes, such as difficulty with language, disorientation, or shifts in mood and behaviour. No two cases are the same. Some patients struggle with everyday tasks, while others maintain independence for longer but show signs of emotional withdrawal or reduced problem-solving ability. This variability means diagnosis and care must be highly personalised, and it’s why a thorough assessment, including cognitive testing, patient history, and input from family, is so essential.”

But, what if, when John first began showing symptoms, clinicians had access to a vast pool of clinical data from hundreds, or even thousands, of other patients to compare against? By looking at his symptoms in the context of this broader dataset, it might be possible to make a more accurate diagnosis and even anticipate how the disease might progress over time by comparing to other, similar cases. 

Working on just this is a team at the UCL Hawkes Institute, a multi-disciplinary centre at the intersection of engineering and health. Within Hawkes is Dr Alex Young, a Principal Research Fellow and Wellcome Career Development Award Fellow. Dr Young has a background in mathematics and computer science, having studied at UCL from undergraduate right the way through to PhD. During this time, she developed an interest in applying algorithms to neurodegenerative disease data and, in collaboration with colleagues in the UCL ‘Progression Of Neurodegenerative Disease’ (POND) group and UCL Dementia Research Centre, she developed an AI machine-learning technique called ‘Subtype and Stage Inference’ or ‘SuStaIn’.

SuStaIn was developed off the back of the idea that diseases like Alzheimer’s are not uniform in their manifestation, progression, or treatment needs, but instead consist of multiple distinct subtypes that evolve differently over time. SuStaIn takes very large and complex datasets like brain scans, symptom scores, or blood tests and can identify distinct subtypes of a disease like Alzheimer’s, mapping out how each subtype develops over time. 

Unlike traditional prediction models, SuStaIn does not assume that everyone is at the same stage of disease progression or has the same type of disease. Plus, because SuStaIn works with cross sectional data and has been adapted to handle missing data, it can build predictions about subtype and progression stage with just a snapshot of data from an individual.

“It’s a bit like constructing a movie from lots of individual photographs” says Dr Alex Young “SuStaIn can compare a snaphot of data to many thousands of other data points and give you predictions of subtypes and progression stages. So, if every patient’s data point is a photo, then the disease progression modelling recreates the movie. But instead of one movie, there are multiple, representing different subtypes of the disease.”

UCL’s POND group have collaborated with centres across UCL and internationally to apply SuStaIn to a variety of neurodegenerative data. In Alzheimer’s disease, a collaboration led by Dr Jake Vogel (McGill University, Montréal, Quebec, Canada) identified four subtypes that have distinct symptoms and are more common than was previously thought. In Multiple Sclerosis, a collaboration led by Dr Arman Eshaghi (UCL Queen Square Institute of Neurology), identified three subtypes that responded differently to treatment and had different rates of disability progression. UCL’s POND group have also applied SuStaIn to frontotemporal dementia and Parkinson’s disease as well as other long-term health conditions such as chronic obstructive pulmonary disease and combined pulmonary fibrosis and emphysema.

Being able to subtype Alzheimer’s disease and predict how it may progress in one patient versus another offers huge promise in terms of more personalised diagnosis, treatment and care. 

Dr Neil Oxtoby, co-founder of the POND group alongside Dr Alex Young and Professor Danny Alexander, is investigating how SuStaIn can support patient care by producing clinician-friendly reports. He’s currently running pilot studies with a memory clinic in Essex, applying SuStaIn to brain scans to generate real-time insights for radiologists.

“I run SuStaIn on each brain scan to produce a personalised, quantitative report for the radiologist. This report details disease subtypes, their implications, prognosis, and potential outcomes for the patient. It’s been an iterative collaboration, with ongoing refinements to make the reports more intuitive and clinically useful. The aim is to eventually develop this into a scalable product for radiology departments, helping to accelerate diagnosis and equip clinicians with personalised insights to support more informed decisions around treatment and care.”

Beyond guiding clinical care decisions, SuStaIn could be particularly valuable in supporting complex or hard-to-make diagnoses.

The early signs of Alzheimer’s disease, such as memory lapses or misplacing items are often hard to spot and common to other disorders. And, while brain scans may be able to detect mild atrophy, it can still be hard to make accurate diagnoses and disease progression predictions on these alone. “SuStaIn can help to improve accuracy and confidence by comparing these tricky cases to the entire dataset” says Dr Oxtoby “it has the potential to really help clinicians with diagnoses, which in turn is beneficial for patients who get more personalised advice when it comes to making care plans.”

As AI tools like SuStaIn become increasingly embedded in clinical practice, fostering close collaboration between clinicians and computer scientists is essential. “It’s a bi-directional process,” explains Consultant Neurologist Professor Jason Warren. “Clinical insights must shape the development and refinement of AI tools to ensure they address the real-world needs of people living with dementia and their support networks. Ongoing evaluation is also key and understanding how these tools perform in practice will help us identify where AI can deliver the greatest benefit.”

Looking ahead, SuStaIn could also play a role in improving clinical trials. Designing trials that stratify patients into meaningful groups based on distinct disease subtypes and progression stages, could enable more targeted interventions, improving trial efficiency, and potentially accelerating the development of effective treatments.

As SuStaIn continues to evolve, it holds promise for accelerating the development of personalised therapies and improving outcomes for patients with complex neurodegenerative conditions like Alzheimer’s.

Related