news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

New Intelligence Report

Editor: Ear Qiao Yang

【New Wisdom Introduction】A Cambridge University study used artificial intelligence to build a machine learning model to accurately predict the development of Alzheimer's disease, with an accuracy rate far exceeding clinical test results, opening up a new path for early intervention in Alzheimer's disease.

If there is one area where artificial intelligence can have an unprecedented positive impact, "healthcare" must be one of the strongest candidates, especially the early diagnosis and treatment of various difficult diseases.

Alzheimer's disease is one of them.

With the progress of population aging, it is like an "eraser" appears in the brains of more and more elderly people. Their memories are gradually erased, they forget the past, and forget their closest family members, which seriously affects the quality of life of the elderly group.

Currently, more than 55 million people worldwide suffer from dementia, with nearly 10 million new cases each year.

The most common type of dementia is Alzheimer's disease, accounting for 60%-70% of all cases. The number of people with dementia is expected to nearly triple in the next 50 years.

Globally, the disease cost healthcare systems $1.3 trillion in 2019 alone.

With Alzheimer's disease, early detection is crucial because that's when treatment is likely to be most effective.

However, without the use of invasive or expensive tests such as positron emission tomography (PET) or lumbar puncture (which are not available in all medical facilities), early diagnosis and prognosis of dementia may be inaccurate.

As a result, up to a third of patients may be misdiagnosed, while others may be diagnosed too late to receive effective treatment.

In addition, for the person suffering from the disease, suffering from the disease may trigger depression and anxiety, not to mention the helpless and desperate relatives around them, who witness the patient gradually lose memory, become mentally ill, and have emotional breakdowns, but they are unable to do anything about it.

However, the development of AI technology has brought some hope to this situation.

Researchers at the University of Cambridge's Department of Psychology have developed a new artificial intelligence model that outperforms current clinical tests in predicting the progression of Alzheimer's disease.

The team used cognitive tests and MRI scans to predict whether people with mild cognitive impairment would develop Alzheimer's disease, while a machine learning model accurately predicted Alzheimer's progression four out of five times using non-invasive data.

Paper address: https://www.thelancet.com/action/showPdf?pii=S2589-5370%2824%2900304-3

The study, also published in the journal eClinical Medicine (published by The Lancet), suggests that the use of AI-based diagnostics could allow for early intervention and reduce reliance on later, expensive diagnostic procedures.

Accurately predicting the progression of Alzheimer's disease

PPM Model

The team built their model using cognitive tests and MRI scans collected by a US research team from 400 patients with grey matter atrophy, the death of nerve cells in the brain.

In order to bridge the gap between artificial intelligence and clinical translation, the paper built a powerful and explainable clinical artificial intelligence tool based on the PPM model (predictive prognostic model).

The model goes beyond binary classification approaches and can predict individuals in the early stages of the disease (mild cognitive impairment, MCI), those in the presymptomatic stage (cognitively normal, CN), and those with confirmed Alzheimer's disease (AD).

PPM introduces a trajectory modeling approach to reliably predict future disease trends from routinely collected multimodal data.

Furthermore, the use of multiple types of patient data, including generalization from research cohorts to real-world patient data, can enhance clinical utility and the potential for adoption in healthcare.

PPM uses the GMLVQ framework of ensemble learning, which is able to combine data from multiple modalities (rather than considering a single data type) to more accurately predict the probability of early symptoms converting to Alzheimer's disease than standard clinical markers (i.e. gray matter atrophy, cognitive decline) or clinical diagnosis.

Harness the power of multimodal data to make predictions from routinely collected non-invasive and low-cost data that may not be as sensitive as biomarkers but are more economical, easy and non-invasive to collect.

This AI-guided, large-model, early-stage forecasting approach has powerful potential:

First, it can predict the health status of patients and reduce healthcare costs, as fewer patients will undergo invasive and expensive diagnostic tests;

Secondly, providing scarce resources to those who need them most can improve the efficiency of medical resource allocation;

Finally, standardizing diagnostics across clinics using large models could reduce inequities in health care.

Sample data collection

The model was tested using a sample of more than 600 participants from the United States, with an additional 900 samples from clinics in the United Kingdom and Singapore.

Data from 600 U.S. samples from the Alzheimer's Disease Neuroimaging Initiative (ADNI) study data cohort were used for PPM training.

An additional 900 samples were used as two independent test datasets for out-of-sample validation, namely the NHS Memory Clinic’s Quantitative Brain Structural and Functional MRI (QMIN-MC) and the National University of Singapore’s Memory Ageing and Cognition Centre dataset (MACC).

These datasets differ in patient demographics and data collection tools, enabling the experiments to simultaneously test the compatibility of the PPM model for different national and regional populations.

Specifically, for ADNI, samples were selected based on specific criteria related to amnestic MCI and Alzheimer's disease.

The MRI data were collected from MRI acquisition sites in the U.S. In contrast, the QMIN-MC and MACC data were collected from representative neurological and psychiatric memory services in the UK and Singapore, respectively.

Therefore, these patient cohorts are generally more reflective of the specific patient profiles encountered in clinical practice than the recruited research cohort (ADNI).

Experimental Results

Test results showed that the algorithm was able to distinguish between people with mild cognitive impairment and those who would develop Alzheimer's disease within three years.

Moreover, using only cognitive tests and MRI scans, it was able to accurately identify 82% of people who would develop Alzheimer's within three years, and correctly identified 81% of people who would not develop Alzheimer's.

Furthermore, it can track the progression of the disease and provide the most appropriate treatment based on the individual circumstances of the case.

The AI ​​model also allowed the researchers to use data from each person’s first visit to separate Alzheimer’s patients into three groups: those whose symptoms remained stable (about 50%), those whose symptoms progressed slowly (about 35%), and those whose symptoms progressed more quickly (about 15%).

These predictions were validated when the researchers tracked the recorded data over a six-year follow-up period. They concluded that the AI ​​solution was three times more accurate than clinical diagnosis in predicting Alzheimer’s progression.

Accurate early prediction is important because it can help identify people who are likely to become ill at an early stage, as well as those who need to be monitored closely because their condition may deteriorate rapidly.

"If we are to tackle the growing health challenges posed by dementia, we will need better tools for early identification and intervention," said Professor Zoe Kourtzi from the Department of Psychology at the University of Cambridge.

“Our vision is to expand the use of AI tools to help clinicians identify diseased groups earlier and use the right treatment. This will also help eliminate the need for unnecessary expensive and invasive diagnostic tests at a time when healthcare resources are under great pressure.”

Early detection is also important to identify misdiagnoses in people who have symptoms such as memory loss but remain stable, as their symptoms may be caused by something other than dementia, such as anxiety or depression.

Dr Ben Underwood, CPFT Honorary Consultant Psychiatrist and Assistant Professor in the Department of Psychiatry at the University of Cambridge, said: “Memory problems are common as we age. In the clinic, not being able to determine whether these are the first signs of dementia can cause a lot of worry for patients.

If we can accurately identify those with stable conditions as not suffering from Alzheimer's disease, this will greatly reduce the psychological pressure on patients.

Professor Kourtzi said: “AI models are only as good as the data they are trained on. To ensure that our model has the potential to be applied in healthcare settings, we trained and tested it not only using data from research cohorts, but also using patient data from the clinic. This suggests that the model will be generalizable to clinical applications.”

Going forward, the research team hopes to expand their model to other types of dementia, such as vascular dementia and frontotemporal dementia, and to use different types of data, such as markers in blood tests.

The research was supported by various institutions including Wellcome, Alzheimer's Research UK, the Royal Society and the National Institute for Health and Care Research Cambridge Biomedical Research Centre.

Early Dementia Diagnosis

In addition to establishing AI predictive models, European universities are actively using artificial intelligence technology to solve the problem of Alzheimer's disease.

With €14 million support from the EU Horizon programme, the AI-Mind project is developing two AI tools that could enable early diagnosis of dementia.

AI-Mind was launched in 2021 and will run until 2026. Its partners include seven European universities, including Aalto University in Finland, Tallinn University in Estonia and Radboud University Medical Center in the Netherlands.

It specifically targets the mild cognitive impairment (MCI) stage, where there are no structural brain defects and intervention is still possible.

To achieve this goal, the 13 partners behind the project are building the AI-Mind Connector and AI-Mind Predictor.

The Connector analyses brain images from EEG to detect early signs that could lead to dementia.

Predictor combines this data with cognitive tests and blood analysis to assess risk of the disease with greater than 95% accuracy.

Both tools will be integrated into the cloud diagnostic platforms currently used by medical professionals.

The ultimate goal of the project is ambitious: to reduce the time to diagnosis from 2 to 5 years to one week. In this way, the hope is to extend the symptom-free period for MCI patients.

Tracking protein clumps

Another use case for AI in the fight against dementia is to deepen our understanding of protein clumps in the body and track how they progress.

For our bodies to function properly, billions of interactions occur between proteins and other molecules within cells.

However, when the interactive reaction process becomes disordered, the proteins aggregate into clumps and further deteriorate, leading to neurological diseases such as Alzheimer's disease.

Researchers at the University of Copenhagen have developed an AI algorithm that can spot protein clumps as small as one billionth of their size in microscope images.

The algorithm can also count clumps, classify them by shape and size, and monitor how they change over time.

This could help scientists understand why these clumps form and lead to the discovery of new drugs and treatments.

According to the team, the tool automates in just minutes a process that would otherwise take researchers weeks to complete.

Moreover, the machine learning algorithm is available for free on the Internet as an open source model.

"As researchers around the world begin to deploy this tool, it will help create a large library of molecules and protein structures relevant to a wide range of diseases and biology," said Nikos Hatzakis, co-author of the study.

References:

https://www.medicaldevice-network.com/news/university-of-cambridge-ai-model-alzheimers/

https://thenextweb.com/news/europe-universities-using-ai-battle-dementia