人工智能可以用于诊断阿尔茨海默症，准确率高达90%

Artificial intelligence is already revolutionizing everything from filmmaking to cybersecurity, and it could also be poised to create major breakthroughs in medicine that have stumped researchers for decades.

The use of A.I. in medicine has been growing in recent years, especially in diagnosing illnesses and diseases. A growing number of doctors already rely on deep learning, a machine learning method modeled on artificial neural networks to learn by example as human brains do, to help detect potentially life-threatening conditions that can be easily missed, such as cancer, heart disease, and even asymptomatic cases of COVID-19.

But the next breakthrough for A.I. in medicine could be in identifying Alzheimer’s, the devastating ailment that causes irreversible cognitive decline and dementia, for which treatment and reliable early detection have eluded medical researchers in the century since the disease’s discovery.

Researchers at Massachusetts General Hospital recently tested deep learning techniques in Alzheimer’s detection, and found that not only was deep learning more accurate than comparative A.I. models that weren’t trained to analyze multiple variables together, it was also able to identify Alzheimer’s cases regardless of factors that usually complicate early-onset detection, such as a patient’s age. The findings were reported in a study published last week in PLOS ONE, a scientific and medical journal.

The researchers trained a deep learning model with tens of thousands of brain scan images collected from over 10,000 people, both with and without Alzheimer’s disease. The study then tested the model against real-world clinical data of Alzheimer’s diagnoses.

The deep learning model was able to identify Alzheimer’s cases with a 90.2% accuracy rate, around five percentage points higher than the simpler A.I. models that did not rely on the deep learning system. The A.I. model performed better regardless of when and where patients were diagnosed with Alzheimer’s, as well as how old they were at the time.

“This is one of the only studies that used routinely collected brain MRIs to attempt to detect dementia,” Matthew Leming, a research fellow at Massachusetts General Hospital and lead author on the study, said in a statement. “Our results—with cross-site, cross-time, and cross-population generalizability—make a strong case for clinical use of this diagnostic technology.”

A 90% accuracy rate in Alzheimer’s diagnosis would be leaps and bounds ahead of human clinical detection rates, which, according to a 2017 study, stand at 77%.

A.I.’s big medical splash

While A.I.-powered search engines developed by OpenAI, Microsoft, and Google have grabbed most of the headlines about artificial intelligence recently for how they promise to disrupt search and how we work, machine learning could have potentially lifesaving applications in medicine.

More than 7 million people admitted to U.S. emergency rooms every year are diagnosed incorrectly, according to a December study by the Department of Health and Human Services. That study found that almost 3 million ER patients are saddled with adverse effects from a misdiagnosis, while over 370,000 suffer from a permanent disability or death.

Misdiagnosis is an economic burden too, as eliminating incorrect testing and treatments as well as the malpractice lawsuits stemming from misdiagnoses could add up to around $100 billion a year in savings, according to the Society to Improve Diagnosis in Medicine, a nonprofit.

Doctors and physicians have said that A.I. holds significant promise in efforts to improve diagnostic techniques, although many of the same issues with A.I. that have been found elsewhere, such as the potential for factual mistakes and racial biases, have also cropped up in medical research. A literature review of A.I. in medical diagnosis published last year found that the technology has promise in fields including cancer, diabetes, and Alzheimer’s diagnosis, although further research is recommended to improve A.I.’s accuracy in identifying medical issues.

A big role in Alzheimer’s research

But if future research makes A.I. and deep learning more widely used in diagnosis, it may be a game changer for Alzheimer’s, which is one of most difficult diseases to predict and diagnose.

Alzheimer’s is the most common type of dementia among older people, afflicting around 44 million worldwide. But it is only one form of a large family of dementia-related conditions, which can easily be misinterpreted as Alzheimer’s.

A 2017 study of over 900 people found that up to one in four Alzheimer’s patients were misdiagnosed, with a roughly even split between false positives and false negatives. Alzheimer’s proclivity for misdiagnosis largely comes down to how many of its symptoms overlap with other common neurological disorders, including Lewy body or frontotemporal dementia. The chances of misdiagnosis increase with age, according to the American Academy of Neurology, which says that Alzheimer’s disease and other dementing illnesses “may be easily misdiagnosed in the elderly.”

Predicting a patient will come down with Alzheimer’s is no easier than diagnosing it, as over 90% of Alzheimer’s cases are considered “sporadic”—appearing in patients with no family link to the disease. Because of these difficulties, there are almost no reliable early screening models for Alzheimer’s, with most cases being diagnosed after symptoms of brain damage begin to be seen.

Massachusetts General Hospital’s study did not address whether deep learning could help with predicting Alzheimer’s, but other studies seem to suggest A.I. could have an important role to play there too.

An A.I. model developed at the University of Florida was able to tap into electronic health records to predict which patients were at a high risk of developing Alzheimer’s up to five years before a diagnosis, the university announced last week. While the researchers recommended more testing before doctors begin employing A.I. predicting tools, they found that A.I. models could help with early diagnosis and reduce the severity of the disease in the long term.