The virtual doctor will see you now: how algorithms are getting better at diagnosing diseases

BLOG: Heidelberg Laureate Forum

Laureates of mathematics and computer science meet the next generation
Heidelberg Laureate Forum

Shwetak Patel saw a lot of 2020 coming. Not the pandemic itself, but the growing importance of technology in medicine. From smart sensors to diagnosis algorithms, medicine is greatly benefiting from technological advancements, and the pandemic has only accelerated these trends. In some regards (especially the technological ones), our society has proven to be remarkably prepared for the pandemic that hit us.

“The current pandemic is taking us into a place where we’re gonna see how we define healthcare,” said Patel at the Digital Health Hot Topic session at the Virtual HLF this past September — a session he proposed in 2019 — and you could hardly imagine a more opportune hot topic.

What Patel and the other panelists pointed out is that while digital medicine is still a work in progress in many ways, we’re already seeing some tangible results. Patel himself developed a way to measure lung function over a simple phone call and is actively working on other smarter ways to diagnose and manage various conditions — and the technology just keeps on developing.

In 2020, we’ve seen some remarkable breakthroughs in the field of medical algorithms. Let’s look at some of them.

An AI that identifies prostate cancer from biopsies

Credit: Ibex Medical Analytics. Prostate biopsy with cancer probability (blue is low, red is high). The AI accurately detected cancer in this tricky case.

Pattern recognition is one of the things AI is best at, and identifying various diseases is one of the most attractive applications for medical algorithms. Researchers at the University of Pittsburgh created a database of a million biopsy images labeled by experts and then ‘taught’ an AI to detect prostate cancer. They then tested the algorithm on 1,600 images from 100 patients from the University of Pittsburgh Medical Center, who were suspected of prostate cancer.

The algorithm reported a 98% sensitivity (detecting the disease correctly) and a 97% specificity (detecting those without the disease). The algorithm can identify clinically important features such as tumor size and spread. Furthermore, the AI flagged six slides that were not initially noted by doctors, and which turned out to be atypical tumors.

Having an AI capable of diagnosing tumors with the accuracy of an experienced doctor is very useful. Not only can this reduce the workload of doctors (doctors who are already overworked in many parts of the world), but it can serve as a helper to aid diagnosticians and improve their own accuracy even more.

Assessing hemoglobin levels with a simple smartphone app

Over 1.6 billion people worldwide suffer from anemia, a condition oftentimes defined by a decrease of hemoglobin in the blood. Many people are completely unaware that they suffer from this, and are also unlikely to get tested. But now, an app may change all that.

Researchers led by Young Kim, an associate professor of biomedical engineering at Purdue University in West Lafayette, Indiana, developed a smartphone app with an algorithm that computes hemoglobin from a photo of the inner eyelid. So far, the accuracy of the test is within 5-10% of the lab-measured value.

While this still won’t eliminate the need for clinical tests, it can serve as a preclinical survey, or as a way to indicate who is at greater risk for anemia — with tools that are already at our disposal. The approach is now being tested on a larger sample size of a more varied population to see if it maintains accuracy.

Doctor-AI teams boost skin cancer diagnosis rates

Two separate studies this year showed how clinicians and diagnosis algorithms can work together to better diagnose skin diseases. First, researchers in Korea developed a deep learning algorithm that accurately classified skin disorders and predicted malignancy (as well as treatment options). The algorithm was compared to various doctors — it performed comparably to dermatology residents, but slightly worse than experienced dermatologists. However, the AI could help inexperienced doctors or the general public get a preclinical assessment, and researchers also noted that it can augment the work of experienced doctors, who performed better in tandem with the AI.

“Our results suggest that our algorithm may serve as an Augmented Intelligence that can empower medical professionals in diagnostic dermatology,” noted lead investigator Jung-Im Na from the Seoul National University, Seoul, Korea.”Rather than AI replacing humans, we expect AI to support humans as Augmented Intelligence to reach diagnoses faster and more accurately.”

In another study, AI-augmented the capacity of clinicians to diagnose skin cancer. The authors found that medical algorithms can “improve diagnostic accuracy over that of either AI or physicians alone, and that the least experienced clinicians gain the most from AI-based support”.

AI can analyze doctor’s notes to distinguish chronic from acute pain

Around 80% of adults experience some lower back pain at some point in their life — often due to a job-related activity. This is often overlooked, both by patients and by doctors. The problem is that the two types of lower back pain (chronic and acute) are very different conditions with different treatments, but they often just get passed in the records as ‘pain’. Now, a new AI has come to untangle the mess in the medical records.

“This study is important because artificial intelligence can potentially more accurately distinguish whether the pain is acute or chronic, which would determine whether a patient should return to normal activities quickly or rest and schedule follow-up visits with a physician,” said Ismail Nabeel, MD, MPH, Associate Professor of Environmental Medicine and Public Health at the Icahn School of Medicine at Mount Sinai. “This study also has implications for diagnosis, treatment, and billing purposes in other musculoskeletal conditions, such as the knee, elbow, and shoulder pain, where the medical codes also do not differentiate by pain level and acuity.”

Ultimately, the approach could be deployed to more than just back pain records and AI could scour medical records, searching for patterns and potential conditions that may have eluded doctors.

A word of caution

Many studies (or more often, press releases) are claiming that AI is as good as doctors or maybe even better at diagnosing diseases from medical images. But before we get carried away, there’s one more study we need to look at.

It was published in the British Medical Journal, one of the more reputed scientific journals, and it warned that these studies are sometimes low-quality or exaggerated.

“At present, many arguably exaggerated claims exist about equivalence with or superiority over clinicians, which presents a risk for patient safety and population health at the societal level, with AI algorithms applied in some cases to millions of patients.”

Overpromising language could mean that valuable technology could ultimately mislead the population. For now at least, the goal of AI is to complement medical activity or at most, to serve as an alternative where medical experts are not available. The not-too-distant-future, however, is looking brighter and brighter.

Posted by

Andrei is a science communicator and a PhD candidate in geophysics. He is the co-founder of ZME Science, where he published over 2,000 articles. Andrei tries to blend two things he loves (science and good stories) to make the world a better place -- one article at a time.


  1. I can imagine a device similar to a photo booth (you know those boxes used for passport photos, etc.) that applies all these diagnostic algorithms described above. I think there is a need for the possibility of anonymous medical advice and appropriately organized AI programs could give that advice. My expectation is that the more controlled the environment in which such an anonymous AI diagnostician operates, the better the results. This is the reason for my proposal of a photo booth-like diagnostic device.

    • A small step for a man, one giant leap for AI-kind.

      In all seriousness though, current medical AIs seem to be more of a complement to doctors and less of a do-it-all algorithm. They’re very niched and specialized for one task and it seems to me that we’re still miles away from generalized AI. This could still be used in some form in underdeveloped areas where medical equipment/specialists are severely lacking, though — and even working imperfectly, it could help save and improve millions of lives.

      I could definitely see smart toilets in our homes gathering data and warning us when something goes wrong or when our eyes suggest a potential anemia in the near future.

      • @Andrei Mihai (citation):

        In all seriousness though, current medical AIs seem to be more of a complement to doctors and less of a do-it-all algorithm. They’re very niched and specialized for one task and it seems to me that we’re still miles away from generalized AI.

        Yes, I agree. But the lack of any consciousness and autonomy of current AI means, that AI remains a tool in the hand of specialists and big companies. I would prefer AI as a friend and helper for all, rather than AI as a helper for the already powerful humans and corporations.

Leave a Reply

E-Mail-Benachrichtigung bei weiteren Kommentaren.
-- Auch möglich: Abo ohne Kommentar. +