AI in the diagnosis and therapy of dementia

Artificial intelligence (AI) promises to bring about groundbreaking advances in medicine as well. In this article, we look at how AI is already being used today and how it will be used in the future for the early detection, diagnosis and therapy of dementia. While AI offers enormous potential, some experts warn of possible dangers. We take a closer look at the opportunities and challenges.

Approximately 1.8 million people in Germany suffer from dementia. In an ageing society, the dimensions of the problem are becoming ever greater: forecasts predict that by 2050, 2.8 million people will be affected. Ever since the first documented description of Alzheimer's disease in 1901, if not before, forgetfulness associated with old age has been the subject of medical research and therapy.

In recent decades, there have been increased efforts to advance from research into the causes to more effective forms of treatment. In this context, it has become increasingly clear how important early diagnosis is for the therapy of dementia-causing diseases. Dementia in old age is not curable, but its progression can be slowed down. The earlier treatment begins, the more quality of life it can maintain.

Utilising the strengths of AI

 Existing AI algorithms have two specific strengths that make them useful for dementia research and therapy. Firstly, AI can analyse large amounts of data, with the results of the analysis tending to improve over time. It can therefore dramatically accelerate research projects that evaluate extensive and complex patient data. Secondly, AI can learn to interact with people: It communicates and adapts the content of its communication to the behaviour of its human counterpart. This makes it an interesting proposition for supporting forms of therapy in which communication is the main focus.

AI to support the diagnosis of dementia

Research institutes have access to extensive patient data of a very wide variety. Much of this data contains clues to incipient dementia or risk factors. However, these clues are often so hidden or vague that even highly qualified people overlook them or interpret them incorrectly. This is where AI can help. Projects are now underway in various countries that have their data sets analysed by artificial intelligence. Some examples:

• Researchers in San Francisco have used AI-based analysis to uncover links between the onset of Alzheimer's disease and other symptoms or illnesses. These links point to common genetic causes that can be identified through follow-up analyses in genetic databases. This way, it is possible to determine up to seven years in advance who is at high risk of developing Alzheimer's.
  
  
• Scientists in Heilbronn are using artificial intelligence to analyse MRI images of the brains of both healthy and diseased individuals to detect changes in certain brain regions that are barely visible to the naked eye. This has also made it possible to identify different types of Alzheimer's cases.
  
  
• A project in Sheffield, for example, is taking a different approach: here, the language of dementia patients is analysed. AI uses voice recordings to identify typical indicators of incipient dementia. In new examinations, the AI then recognises within a few minutes from the language of an examined person whether signs of dementia are showing. The CognoSpeak tool from the University of Sheffield can even be used at home.
  
  
• Researchers at the US Mayo Clinic have developed a system in which artificial intelligence analyses patients' EEG data (brain waves) to identify typical abnormalities.
   

The basic principle of this and other projects is usually similar: the AI scans a data pool and identifies data anomalies that are typical of certain symptoms or diseases. During patient examinations, it is then checked whether there are indications of a typical anomaly in this person, perhaps only in a very slight form – which could indicate the onset of dementia at a very early stage.

AI in the treatment of dementia patients

Early diagnosis of dementia gives patients time to undergo therapies designed to maintain mental abilities for as long as possible. Artificial intelligence can also help here. Its ability to learn and adapt can be used to develop personalised therapies.

For example, a project at the Hamm-Lippstadt University of Applied Sciences is developing algorithms to support cognitive stimulation therapy, specifically for hearing loss associated with dementia. In this form of therapy, patients are encouraged and guided in activities that train cognitive and social skills. AI first analyses collected patient data and transfers it to a training programme. The algorithm then tests which actions can actually be successfully performed by a specific patient and adjusts the training level individually based on the results. In doing so, the AI gradually learns which actions lead to positive results for this patient and which do not. Continuous analysis of the data collected enables a kind of therapy supervision.

The line between dementia therapies and non-medical measures that aim to improve the quality of life of people with dementia in their everyday lives is blurred. The Dementia VoiceBot, which is being developed One of the biggest problems associated with dementia is loneliness, which arises, among other things, from the fact that sufferers can no longer communicate with other people in a satisfactory way. AI instances, on the other hand, can adapt

The University of New South Wales in Australia is therefore testing a ‘digital companion’ – an AI video personality that engages in everyday conversation with dementia patients in a way similar to a video call.

Robots as patient companions for people with dementia

The latest step is the AI-controlled robot. For over a decade, companies in Japan have been trying to market humanoid robots as everyday aids and companions for lonely people. The Lovot from Groove X, for example, was also developed with older people in mind. It learns how to move around safely in an apartment and how to react appropriately to the emotional expressions of its owner – in a sense, it simulates the behaviour of a pet. The camera can be accessed remotely via an app, enabling caregivers to check on a person living alone.

The robot QT, which is being tested at the University of Indiana, is tailored to the needs of dementia patients. QT engages in conversation and adapts to the level of its counterpart. In the future, such robots could to a limited extent take over the role of therapists: as the number of people suffering from dementia grows, it will become increasingly difficult to train enough human therapists. Perhaps robots could fill the gap for less demanding tasks.

AI for the diagnosis and treatment of dementia – blessing or curse?

Is this all a promising future, or does the use of AI in dementia also entail significant risks? There is indeed serious criticism of its use.

• Dehumanisation of the relationship between doctors and patients
  AI can adapt to the behaviour of patients on the outside, but it cannot really understand it. Isn't there a risk here that individual characteristics and needs will be overlooked too often? It is often not apparent to humans how an AI arrives at its decisions. This will become a problem at the latest when these decisions are faulty or harmful. If something goes wrong, it is often impossible to say why and where it originated.
  
  
• Lack of transparency when errors occur
  It is often not apparent to humans how an AI arrives at its decisions. This becomes a problem at the latest when these decisions are faulty or harmful. If something goes wrong, it is often impossible to say why and where it came from.
  
  
• Excessive trust in technology
  The use of complex AI models whose decision-making processes can no longer be understood can also lead to a false sense of security. Then it is not recognised in time when the AI is following the wrong approach.
  
  
• Data protection
  AI systems process sensitive health data. As is so often the case with AI, the question arises as to whether this data is sufficiently protected – and who is allowed to use this data and for what.
  
  
• Environmental impact
  AI requires considerable computing resources and therefore high-performance processors. AI data centres consume a particularly large amount of energy.
  
  
• Discrimination against patients from non-Western or poorer countries
  The patient data used to train AI comes mainly from Europe and North America; for people from other regions, it is sometimes less meaningful. For example, language analysis tools trained with data from English-speaking patients are of little help to people with a different mother tongue. However, AI systems are expensive and therefore cannot be developed everywhere in the world. AI diagnoses currently mainly help people who already have better healthcare.
   

If we want to use AI as a promising engine for improved medical diagnostics and care, we have to find answers to all these reservations.

By Alexa Brandt and Thorsten Kleinschmidt