When the world gets closer.

We help you see farther.

Sign up to our expressly international daily newsletter.

Future

Data, Selfies, Prevention: How AI Is Transforming Healthcare

From testing for COVID through WhatsApp to taking selfies to check heart risks, AI programs are being used in Argentina to complement early-stage diagnoses. The technologies are in their early stages but are able to detect what the human eye might miss.

brain MRI

In Argentina, AI technologies are used to complement early-stage diagnoses

Lautaro García Alonso

BUENOS AIRES —The World Health Organization (WHO) estimates that every year 138 million patients suffer from medical misdiagnoses that prove fatal in 2.6 million cases. In the United States, medical errors relating to misuse of pharmaceutical products or misdiagnosis were the third cause of deaths there in 2015.

All this proves that medicine is not infallible, and even specialists can go wrong. The daily performance of all doctors is subject to factors like stress, overwork or exhaustion (they sometimes work 24 hours straight). In this context, technological advances of recent years may bring some good news. Artificial Intelligence (AI) has brought innovations that boost diagnosis and even detect conditions invisible to the naked eye.


In Argentina, AI application in healthcare is in its initial stages but showing promising developments.

Testing for COVID through WhatsApp

Recently, the Buenos Aires city government launched IATos, a program meant to boost testing for the coronavirus. The tool allows you to record a cough on WhatsApp and send it on through Boti (the city's chat bot) for analysis by the IATos program. If the sound matches patterns of positive cases, the program recommends you get a COVID-19 test.

IATos works on the basis of a neural AI network that can analyze sound, breathing and cough sounds.

To train the web's recognition system, the sounds of 140,000 COVID-infected or negative patients were gathered through Boti (through PCR tests) from mid-2020. It is the biggest database of its kind in the world and freely accessible to all.

The program has raised privacy concerns.

Current predictions made by IATos are 86% accurate, though as the city's Undersecretary for the Intelligent City Agustín Suárez told Clarín, "this is not a detection test but another element weighing on whether or not to test yourself." That means Boti's guideline must be confirmed by a nose swab.

The program has raised privacy concerns. Suárez says "the recordings are made anonymously and there is no collection of personal data. The system only analyzes specific coughing patterns." On the software's access to your device's microphone, Suárez says only information received through the WhatsApp sound system is processed, adding there was no need to activate more permits or download any other application.

Coronary and diabetes risks

Julio (65) went to a cardiologist after feeling unwell. He sensed it could be a blocked artery. And yet after filling a routine questionnaire and completing tests, he was told he was fine. A few years on, Julio has recovered from heart surgery, which he underwent for complications that came after he took his tests. He says that had there been a tool to predict his risk of a heart attack, it might have persuaded his doctor that he could be sick.

Gustavo Daquarti, a cardiologist, heads the AI unit at ÜMA Salud, a digital health platform. He told Clarín he became interested in AI "when I worked on diagnoses through imaging and began noticing that the algorithm suggested something different to what I was thinking. I then checked and found I had been wrong and AI was right."

He says patients typically visit the doctor when symptoms appear, which is why ÜMA Salud has focused on developing tech tools to detect pathologies early. In September 2021, they launched a free tool on their webpage that uses an algorithm based on deep learning of neural networks to predict the risks of heart disease or Type 2 diabetes. For that the patient must simply register, fill in a brief questionnaire, and send a face picture (a selfie), which can be taken with any device provided there is light and the face is shown clearly.

The system analyses the image in seconds and determines the person's risk levels for those illnesses. The algorithm here analyses certain face patterns (formation of certain wrinkles on the ear, small fat deposits on eyelids etc.) that may be associated with very early signs of a possible pathology. These patterns are often invisible to the eye. Daquarti says a pilot test carried on 1,500 images (500 of diabetic patients, 500 of heart patients and 500 healthy individuals) had a 71% level of accuracy for heart disease and 72% for diabetes. Today a cardiac stress test has a 70% accuracy level.

More than 8,000 people have used the platform, though Daquarti stresses its findings are indicative, "not a definitive diagnosis." Again, clinical tests were needed for confirmation.

AI as a useful ally

transparent skull

Deep-learning algorithms are also being used in neurological studies

Unsplash/Jesse Orrico

95% accuracy rate

Mauricio Farez, a neurologist, currently heads the CIEN research center into nerve and immune system diseases at FLENI, a neurology center and clinic. In 2017 he co-founded Entelai, a startup developing AI in healthcare. One of Entelai's focuses has been to incorporate AI into diagnostic imaging. Its Entelai Pic product is a tool based on deep-learning algorithms being used in neurological studies. It is the first AI-based medical software to be approved by the government drug regulation agency.

On the one hand it is used in the early detection of demyelinating diseases like Multiple Sclerosis, which requires extremely precise magnetic resonance. AI allows for detection of changes in the brain volume and certain lesions that are very difficult to spot visually.

Accuracy in detecting lung affections is above those of even the most specialized physicians.

Through algorithms, the tool can also measure brain volume and those sections particularly relevant in diagnosing brain degenerative diseases like Parkinson's or Alzheimer. Farez says it is very difficult for a medic to view a three-dimensional reconstruction of the brain and compare it to "what is considered normal, unless the alterations are very severe. Today, thanks to AI, we can precisely determine the patient's brain volume and contrast the value with a database of thousands of cases."

Entelai Pic also allows thorax radiographies, which made it of particular use in the early months of the pandemic when the coronavirus provoked the greatest number of pneumonias. Its accuracy in detecting lung affections is above those of even the most specialized physicians.

Farez says AI is as effective in the diagnosis process as the best specialists, but, he adds, scientific evidence shows that a combination of both types of diagnosis beats either one separately. He calls AI an "ally in taking better decisions."

The firm has had 250,000 consultations using Entelai Pic, and hopes by the end of 2022 to have a million images analyzed with AI. While the firm has so far worked with clinics and hospitals, it wants to launch a platform for private users by the end of March this year.

In this case, it will be a tool allowing patients to analyze moles through pictures taken on their cellphones, for early detection of possible melanomas. Tests so far put the tool's accuracy rate at 95%.

Deep learning and breast cancer

When Carolina Mondino (25) decided to study bioengineering, she knew this was to improve people's lives. For her final project, she designed an AI-based tool to assist medics in detecting breast cancer.

She designed an algorithm that can classify tumors observed in mammograms according to type and severity of the lesion, and thus determine their malignancy. The system is currently 90% effective in classifying tumor types and has a 70% accuracy rate in determining malignancy. She too hopes the program, initially conceived for medical use, can later be used by patients themselves.

Entelai Pic also has an algorithm to detect tumors in mammograms, with a 93% accuracy rate. Farez says the algorithm used was trained using a compilation of one million images. One of AI's great advantages, he added, was speed and the ability to cut down physicians' work and delays, which may be crucial in cancer treatment.

Promising developments

Both ÜMA Salud and Entelai are working on the concept of assisted diagnosis, or an AI tool that makes a preliminary diagnosis through questions to the patient. ÜMA Salud's algorithm was trained using thousands of clinical histories and has an 80% accuracy rate. Daquarti says it is still being developed. Entelai's product, Entelai Doc, includes referral to specialists, automated patient monitoring and an alarm system for urgent consultations.

Other applications remain at the investigative stage, but may prove decisive later. One is in the use of AI to drastically cut down time needed to differentiate pluripotent stem cells. This, says Ariel Waisman, a CONICET researcher in this project, will have implications in regenerative treatments for a range of human organs. He says "pluripotent stem cells are the most powerful type of stem cells, as they can 'convert' (differentiate) into any cell type in the adult organism."

The Argentine Science Ministry is also testing AI-based algorithms designed to prevent outbreaks of COVID and other contagious diseases in the country.

You've reached your monthly limit of free articles.
To read the full article, please subscribe.
Get unlimited access. Support Worldcrunch's unique mission:
  • Exclusive coverage from the world's top sources, in English for the first time.
  • Insights from the widest range of perspectives, languages and countries
  • $2.90/month or $19.90/year. No hidden charges. Cancel anytime.
Already a subscriber? Log in

When the world gets closer, we help you see farther

Sign up to our expressly international daily newsletter!
Coronavirus

Will China's Zero COVID Ever End?

Too much has been put in to the state-sponsored truth that minimal spread of the virus is the at-all-cost objective. But if the Chinese economy continues to suffer, Xi Jinping may have no choice but to second guess himself.

COVID testing in Guiyang, China

Cfoto/DDP via ZUMA
Deng Yuwen

The tragic bus accident in Guiyang last month — in which 27 people being sent to quarantine were killed — was one of the worst examples of collateral damage since the COVID-19 pandemic began in China nearly three years ago. While the crash can ultimately be traced back to bad government policy, the local authorities did not register it as a Zero COVID related casualty. It was, for them, a simple traffic accident.

The officials in the southern Chinese province of Guizhou, of course, had no alternative. Drawing a link between the deadly crash and the strict policy of Zero COVID, touted by President Xi Jinping, would have revealed the absurdity of the government's choices.

Keep reading...Show less

When the world gets closer, we help you see farther

Sign up to our expressly international daily newsletter!
You've reached your monthly limit of free articles.
To read the full article, please subscribe.
Get unlimited access. Support Worldcrunch's unique mission:
  • Exclusive coverage from the world's top sources, in English for the first time.
  • Insights from the widest range of perspectives, languages and countries
  • $2.90/month or $19.90/year. No hidden charges. Cancel anytime.
Already a subscriber? Log in
THE LATEST
FOCUS
TRENDING TOPICS

Central to the tragic absurdity of this war is the question of language. Vladimir Putin has repeated that protecting ethnic Russians and the Russian-speaking populations of Ukraine was a driving motivation for his invasion.

Yet one month on, a quick look at the map shows that many of the worst-hit cities are those where Russian is the predominant language: Kharkiv, Odesa, Kherson.

Watch VideoShow less
MOST READ