Innovation Monitor: AI’s impact on blindness, PTSD and cancer
Welcome to this week’s Innovation Monitor.
“The 21st-century economy has been a two-decade series of punches in the gut.” — Neil Irwin
In the past few decades, economic writers have defaulted to a subset of themes, ranging from “ominous to gloomy to terrifying,” remarks NYT’s senior economic correspondent Neil Irwin. He chronicles the last twenty years as a “mild recession… followed by a weak recovery followed by a financial crisis followed by another weak recovery followed by a pandemic-induced collapse.” Ominous, gloomy, terrifying.
But Irwin is finally, cautiously optimistic. In his excellent long-read he lists 17 reasons why… which we recommend reading in full. But today we’re focusing on Irwin’s first points, that the Salow paradox may have run its course:
“In 1987, the economist Robert Solow said, ‘You can see the computer age everywhere but in the productivity statistics.’ Companies were making great use of rapid improvements in computing power, but the overall economy wasn’t really becoming more productive. This analysis was right until it was wrong. Starting around the mid-1990s, technological innovations in supply chain management and factory production enabled companies to squeeze more economic output out of every hour of work and dollar of capital spending.”
Shake and shake the ketchup bottle. First none will come and then a lot’ll. And what “ketchup” is more relevant today than AI? In this newsletter, we’ll take a look at how the past decade of progress in AI and other emerging tech has led to some breakthrough technology in healthcare.
Despite fears of misdiagnosis, bias, and disparate accuracy outside the lab, both physicians and technologists are optimistic for the coming decade — that’s saying something.
This week, we’re exploring the growing field of restorative AI, and how emerging technologies can help address blindness, PTSD, cancer detection, and Sepsis diagnoses.
Finally, for my Asian American sisters and brothers, this has been a deeply painful week and year for us. In NYC this past year, there’s been an over 1900% increase in reported hate crimes against Asian identifying individuals. It takes all of us to stop Asian hate and be supportive of each other and our communities.
As always, stay safe & thank you for reading, and if you were forwarded this email, you can easily sign up here!
Erica Matsumoto Blindness Ophthalmologists have gotten very good at treating diabetic retinopathy — the leading cause of blindness, with 24,000 adult cases per year in the US — but they have trouble detecting it early. Regular screening is vital, but out of 400M people in the world with diabetes, about half get their yearly eye checkups.
It’s not just people being lazy about their appointments — there is a shortage of ophthalmologists, writes Nature’s Sandeep Ravindran, especially in low- to middle-income countries. Even high-income countries are strained. Ravindran covers retinal specialist and computer scientist Michael Abràmoff’s IDx-DR AI system, developed over the last two decades to detect more-than-mild cases of diabetic retinopathy.
Unlike, say, the story of Google’s medical AI at work in Thailand — a quintessential example of what can go wrong with high-accuracy AI models in the real world — IDx-DR is the first device to be approved by the FDA “to provide a screening decision without the need for a clinician.”
AI isn’t just employed in preventative measures — last year the Google Glass team experimented with using the glasses to “extract visual information from images of people, belongings, and public transport, and then [speak] about them out loud.” Microsoft is working on Project Tokyo — a modified HoloLens that helps blind or low vision people identify their surroundings and develop social interaction skills.
What do the coming decades hold? One possibility is invasive BCIs that hook directly into the brain and feed electrical signals. MIT Tech Review has a moving story of 57-year-old Bernardeta Gómez getting a brain-to-camera implant to see basic shapes.
PTSD In various capacities, AI has been used to help patients with PTSD, which impacts 8M adults in the US. In a 2019 study led by Charles R. Marmar, professor and chair of the Department of Psychiatry at the NYU School of Medicine, an AI system achieved 89% accuracy determining which participants had PTSD just by analyzing voices.
“The random forest program linked patterns of specific voice features with PTSD, including less clear speech and a lifeless, metallic tone, both of which had long been reported anecdotally as helpful in diagnosis. While the current study did not explore the disease mechanisms behind PTSD, the theory is that traumatic events change brain circuits that process emotion and muscle tone, which affects a person’s voice.”
Last year, an AI system performed a much simpler task — cutting six of the 20 questions used to diagnose PTSD. “Researchers built a machine learning model that learned how strongly different terms in the diagnostic predicted PTSD diagnosis. This enabled the team to identify which items had weak associations and could be cut, while maintaining at least 90 percent accuracy.”
A few years ago researchers paired hundreds of patients with Ellie, an AI therapist designed to listen to PTSD patients and offer helpful feedback and follow-up questions: “For example, Ellie not only knows how to perform sympathetic gestures, like nodding, smiling, or quietly uttering “mhm” when listening to a sensitive story — she knows when to perform them.”
Cancer A March 2020 story in Nature on AI-based cancer detection demonstrated how useful diagnostic AI can be as a second opinion, facilitating pathologists’ hunches instead of replacing them.
A young girl arrived at NYU Langone Health for routine tests on her recurrent cancer. “With this diagnosis, the girl would begin a specific course of radiotherapy and chemotherapy. But just as neuropathologist Matija Snuderl was about to sign off on the diagnosis and set her on that treatment path, he hesitated.”
Snuderl thought the biopsy was unusual, and fed the data to an AI system for a second opinion. The tumor came back as a completely different type, “and called for a different drug and radiation treatment plan. Treatment for the wrong cancer could have ill effects without actually destroying the cancer.”
AI has also been used to catch cancer early on. MIT computer scientist Regina Barzilay collected 89,000 mammograms from nearly 40,000 women to train an early-detection algorithm.
“The computer put 31% of the women who eventually developed breast cancer into the highest risk group. But the standard Tyrer–Cuzick model that physicians use to estimate risk… placed only 18% in that group, even when physicians added measurements of breast density from mammograms to the model.” Sepsis Sepsis — full-body inflammation that could lead to organ failure — is notoriously difficult to diagnose early, making it one of the leading causes of hospital deaths. To help spot signs early, a team at Duke University launched Sepsis Watch in 2018. The product of over three years of development — including the analysis of 32M data points — the system scores patients for the likelihood of sepsis onset.
If a patient seems at risk, the system sends an alert and a doctor confirms the diagnosis. After two years online at the emergency department of the Duke University Health System, the consensus is that Sepsis Watch works. “It has dramatically reduced sepsis-induced patient deaths and is now part of a federally registered clinical trial expected to share its results in 2021.” This Week in Business History March 15th, 1662: The first modern public transport is invented in Paris.
Designed by mathematician/philosopher/scientist Blaise Pascal in 1661 and financed by the royal court, the system started with seven horse-drawn vehicles running along regular routes which could each carry six to eight passengers. The system was called Carrosses à cinq sols, or five-penny coaches, and unfortunately the novelty wore off quickly, and the concept of organized public transportation would not return again until the mid-1800s.
This email was sent to <<Email Address>>
why did I get this? unsubscribe from this list update subscription preferences
NYC Media Lab · 370 Jay Street, 3rd floor · Brooklyn, New York 11201 · USA