Using AI to See the Hidden Signs of Breast Cancer

Summary

WSJ Tech News Briefing — Using AI to See the Hidden Signs of Breast Cancer

Date: October 3, 2025
Host: Peter Ciampelli
Guests: Nicole Nguyen (WSJ Personal Tech Columnist), Brianna Abbott (WSJ Reporter), Katie Dayton (WSJ Contributor)

---

Episode Overview

This episode explores two rapidly evolving realms of AI technology: smarter automation in the home and groundbreaking advances in medical imaging for cancer risk prediction. The first half focuses on major upgrades headed to consumer smart home devices like security cameras and assistants. The second half dives into new AI tools that can predict the risk of breast cancer from a single mammogram, potentially transforming clinical screening and diagnostics.

---

Smarter, AI-Driven Smart Home Tech

Guest: Nicole Nguyen, WSJ Personal Tech Columnist

Key Discussion Points

• Major AI Upgrades for Cameras

  • Security cameras (Nest, Ring) are about to get much smarter, thanks to AI integration.
  • New abilities include understanding scene context, so notifications are more useful and less generic.
  • Example: Google’s Gemini allows specific queries like, “What’s eating my plants?” The AI reviews footage and can identify, for instance, rabbits in the garden.
  • Amazon leverages its whole network of Ring cameras for tasks like locating missing pets, using AI to identify animals across multiple devices.

• Networked AI for Community Problem-Solving

  • Ring owner can initiate a search for a lost dog, with local cameras looking for the pet and owners able to opt-in to share relevant footage.
  • This technology has potential broader applications, e.g., for finding missing persons who have dementia.

• Quote Highlight

  • “In this demo that I saw, [Gemini] surfaced a video clip of two rabbits munching on grass.” — Nicole Nguyen [02:09]
  • “You can trigger a search party in the app [Ring] which will look for your pup using AI in footage captured by nearby cameras.” — Nicole Nguyen [02:37]

• Privacy and Security Concerns

  • Previous controversy involving Ring’s sharing of footage with law enforcement has made opt-in sharing crucial for these new features.

  • Owners remain in full control over whether to share footage triggered by AI events.

    • “The head of Ring told us it was really important for Ring camera owners [...] to be able to opt in or out of sharing.” — Nicole Nguyen [03:29]

• Smart Home 2.0: Natural Language Automations

  • Early smart home tech was largely timer-based or required complex, multi-app routines.

  • The new wave allows users to trigger complex automations with simple statements, e.g. “Make me feel safer,” prompting actions like checking doors, windows, and toggling lights automatically.

    • “A lot of those automations and routines will be triggered by natural language.” — Nicole Nguyen [04:20]

Timestamps:

• [01:24] — AI changes coming to smart security cameras
• [03:16] — Privacy and opt-in controls discussion
• [03:56] — Comparing “Smart Home 1.0” vs. “2.0”
• [04:54] — Segment transition

---

Fighting Breast Cancer with AI: A Clinical Frontier

Guests:

• Brianna Abbott, Wall Street Journal Reporter
• Interviewed by Katie Dayton

Key Discussion Points

• Traditional Breast Cancer Risk Assessments

  • Typically use quiz-based calculators including factors like age, ethnicity, breast density, and menopausal status.

  • Classifies a 3% five-year risk as high, or a 20%+ lifetime risk as high.

    • “They quiz women on their age, their ethnicity, breast density, menopausal status, like things that are well known to raise or lower your breast cancer risk.” — Brianna Abbott [06:10]

• Limitations of Current Tools

  • Existing calculators are only accurate about 60% of the time.

  • AI models already surpass that, achieving 70-80% accuracy in discriminating future cancer cases, representing a meaningful jump in predictive ability (but are not perfect).

    • “For the AI models [...] that increases to 70% of the time or more.” — Brianna Abbott [06:54]
    • “That is a pretty big jump. But it’s still obviously not perfect.” — Brianna Abbott [07:14]

• AI’s Unique Capabilities

  • Human radiologists cannot reliably use a single mammogram to identify future cancer risk.

  • AI can analyze vast sets of scans linked to later outcomes, offering a personalized risk score from one image.

    • “She can’t look at a mammogram and predict a woman’s five year risk of breast cancer with any accuracy. But that is the point where the AI is able to step in.” — Brianna Abbott [07:40]

• Potential Downsides and Challenges

  • Over-flagging of lower-risk women, leading to unnecessary tests or even preventative treatments.

  • Questions remain about bias: AIs trained on one population might not generalize well; e.g., white vs. Black patient populations.

  • Regulatory gap: FDA does not require clinical trials for these AI tools, so some in medicine want more rigorous evidence.

    • “There’s a risk that you flag a lot of women who are lower risk [...] when they didn’t necessarily have to do that.” — Brianna Abbott [08:24]
    • “If you train it on a data set of predominantly white patients, how well is it going to work in a hospital that has predominantly black patients?” — Brianna Abbott [08:44]

• Regulatory and Clinical Reality

  • No clinical trials required for approval (unlike medicines), but they're desired by many radiologists to prove real impacts like reduced mortality or late-stage disease.

  • Fast pace of AI makes traditional clinical trial timelines challenging — by the end of a five-year trial, the model tested may be outdated.

    • “The way that the Food and Drug Administration works is that these AI models don’t actually need to go through clinical trials to get approved.” — Brianna Abbott [09:21]
    • “The bar that these models need to clear is not as high as a pharmaceutical drug.” — Brianna Abbott [09:31]

• Deployment Timeline

  • Clarity AI received FDA approval earlier in 2025; roll-out in clinics is expected by end of year, others (Deep Health) close behind.

  • Applications to other cancers and genetic conditions are in earlier stages.

    • “You could be seeing something like this at your routine mammogram somewhere sooner than you think.” — Brianna Abbott [10:39]

Timestamps:

• [06:08] — Overview of current breast cancer risk calculators
• [06:45] — Comparing AI accuracy to existing methods
• [07:36] — AI’s ability to predict from a single scan
• [08:20] — Concerns about false positives and population bias
• [09:21] — Regulatory discussion: FDA, clinical trials
• [10:30] — First approvals and what’s next

---

Notable Quotes

• “In this new wave of smart home stuff, which we can call Smart Home 2.0, a lot of those automations and routines will be triggered by natural language.”
  — Nicole Nguyen [04:20]

• “The algorithm can basically come up with a way to spit out a risk score from a single scan.”
  — Brianna Abbott [07:55]

• “Some of the AI companies, their response to that is why would we be held to a higher standard that these quiz calculators are not?”
  — Brianna Abbott [10:06]

---

Summary

This episode captures two pivotal trends at the intersection of AI and everyday life: smarter, context-aware smart homes and a paradigm shift in disease risk assessment via AI-analyzed imaging. The discussion underscores both the promise—more meaningful alerts and detection, earlier interventions for disease—and the pitfalls, such as privacy, regulatory, and algorithmic fairness concerns. Real-world deployment of AI tools for breast cancer risk in the clinic is imminent, heralding a future where simple scans could reveal hidden medical threats years in advance.

Transcript

[Veeam/GE Vernova Advertiser]

Cybercriminals. Count on chaos. Count on Veeam to stop them fast. Partner with the global leader in data resilience and get 24. 7 ransomware support for your business with Veeam. It's all good. Keep your business [e-mail] welcome to Tech News Briefing.

[Peter Ciampelli]

It's Friday, October 3rd. I'm Peter Ciampelli for the Wall Street Journal. Imagine your dog runs off and all of your neighbors ring cameras automatically start looking for it. Those types of advancements are to smart home technology, from smarter security cameras to AI speakers. And then researchers are developing AI tools that can predict a patient's risk of breast cancer just from a routine mammogram. So how does it compete against other kinds of screening? We'll find out more about some new health tech that could be introduced in clinics as soon as this year. But first, after years of stagnation, smart home technology could be getting a lot more powerful, thanks to some updated features set to roll out from Google later this month and ones that are already rolling out from Amazon. Wall Street Journal personal tech columnist Nicole Nguyen is here to tell us more. So Nicole, it sounds like there are a few categories of smart home tech where there's going to be upgrades. What kind of changes are coming for security cameras like Nest or Ring?

[Nicole Nguyen]

The biggest changes will be to the way you interact with security cameras, which are probably the most popular smart home device outside of smart speakers. AI has an incredible ability to understand what's going on in images or video. And this is really helpful for security cameras because they're always rolling and they're always recording. But to find an action, you either have to scrub through hours of video or you get a notification. That's pretty vague. It's like motion detected. And so what Google's Gemini and Amazon's Alexa plus that's the revamped version of Alexa that Amazon has started rolling out earlier this year can do is understand what's happening and give you better notifications. So for example, the Nest cam update that just went out earlier this month is that you can ask Google Gemini, what's eating my plants? And if there's a camera that's pointed at your garden, Gemini will spring into action. Look through your footage and in this demo that I saw, it surfaced a video clip of two rabbits munching on grass. Amazon is taking this AI powered insight technology a step further by leveraging its massive network of outdoor cameras. So this is ring, video doorbells, outdoor floodlights, that kind of thing to look for missing dogs. So if you have a Ring app and your dog goes missing, you can trigger a search party in the app which will look for your pup using AI in footage captured by nearby cameras. And if that nearby camera thinks that it sees your dog, that device's owner will get the option to share that footage with you, the dog's family. The head of Ring mentioned that you can imagine people who go missing who have dementia. There could be other interesting use cases for this kind of technology.

[Peter Ciampelli]

Are there any safety or data privacy concerns from those inside or outside the company about these upgrades to smart home tech?

[Nicole Nguyen]

People have had concerns about Ring cameras because of issues involving videos being shared with law enforcement. And so in designing this feature, the head of Ring told us it was really important for Ring camera owners whose devices allegedly saw a dog for them to be able to opt in or out of sharing.

[Peter Ciampelli]

And then how will the advancements in automation differ with this new tech versus what smart home tech could already do in terms of existing abilities to program things like timed light dimmers or auto locking doors?

[Nicole Nguyen]

Yeah, so I'd say that Smart Home 1.0, the first generation of smart home products, was very much like you basically use your smart speaker for super basic stuff like setting a timer for 30 seconds or playing your Discover weekly on Spotify. And then the true smart home stuff like routines that involve your thermostat and lights and your garage door really was for tinkerers to set up these very complicated routines that involved like fussy installation and settings across multiple apps. And in this new wave of smart home stuff, which we can call Smart Home 2.0, a lot of those automations and routines will be triggered by natural language. So in a demo that I saw for Gemini for home uttering make me feel safer, Gemini understood to okay, I'll check that the doors are locked and that the windows are closed because there's sensors on the windows. And if you're not home toggling the lights on and off.

[Peter Ciampelli]

That was our personal tech columnist, Nicole Nguyen. Coming up, companies like Clarity and Deep Health are designing AI that can look at medical scans and see details that are too small or complex for the human eye. Is it ready to be used in the real world? That's after the break.

[Veeam/GE Vernova Advertiser]

Optimism isn't sunshine and rainbows. It's fixing things, changing the way we fix things. It's running the world on smarter energy. Because if optimism never stops, then change can't either. GE Vernova the energy of change.

[Peter Ciampelli]

Artificial intelligence is making its way into more and more avenues of healthcare. One that's right around the corner. Researchers and companies are designing AI tools that are trained on hundreds of thousands of past mammograms, then matched with those patients status five years later to be able to predict signs of breast cancer. The Wall Street Journal's Katie Dayton talked with reporter Brianna Abbott to find out more.

[Katie Dayton]

So Brianna, before we get into the AI of it all here, how are doctors typically assessing a woman's risk of developing breast cancer without AI?

[Brianna Abbott]

The most standard common method to sort of gauge a woman's level of breast cancer risk are these sort of quiz based calculators. And so what they do is they quiz women on their age, their ethnicity, breast density, menopausal status, like things that are well known to raise or lower sort of your breast cancer risk. And from there it calculates a score. And as of right now, a score of about like a 3% risk in the next five years is considered high, which is or a lifetime risk of 20% or more.

[Katie Dayton]

And what are the limitations of that method and how does AI help bridge them?

[Brianna Abbott]

So the limitations are that the tests aren't bad, but they're not that great in terms of accuracy. So right now, around 60% of the time or more, the calculators will correctly give a higher risk score to a woman who will develop cancer over a woman who won't develop cancer. And for the AI models and that increases to 70% of the time or more depending on the group or the specific model. Some of the studies show 80, some show 75, that sort of thing, but they are consistently higher. And that is a pretty big jump. But it's still obviously not perfect.

[Katie Dayton]

So from your reporting, doctors now they have the scans, but they're not able to tell from a scan taken today that that person will develop cancer in the future. From the picture. And is it the case that AI can do that?

[Brianna Abbott]

Yeah, so that is the goal of these newer AI models. Like I was sort of talking with radiologists, one of them essentially said that she can't look at a mammogram and predict a woman's five year risk of breast cancer with any accuracy. But that is the point where sort of the AI is able to step in. If you give the a mammogram and say in five years this woman did or did not develop breast cancer, the algorithm can basically come up with a way to spit out a risk score from a single scan.

[Katie Dayton]

And so if they're predicting the chance of breast cancer in patients at a higher rate than traditional calculator assessments, are there any downsides to that, especially as we're talking about risk and not the presence of cancer itself.

[Brianna Abbott]

Like we said, these are not perfect. And there's a risk that you flag a lot of women who are lower risk breast cancer and then they go through extra screening or, or some of these preventative drugs when they didn't necessarily have to do that. So that's a challenge. And then obviously there's lots of concern as some of these get rolled out on the market pretty quickly. There's running concern that a models that are trained based on a data set of certain populations might not work as well. Like if you train it on a data set of predominantly white patients, how well is it going to work in a hospital that has predominantly black patients? So you want to make sure that these things cross over and work equally as well in all different patient demographics.

[Katie Dayton]

You mention in your story the tension between the need to conduct clinical trials of screenings like this one, which take years to complete, and the speed in which AI is developing. How have you seen the industry grapple with that disparity in timing?

[Brianna Abbott]

That's a tricky thing. And to be clear, the way that the Food and Drug Administration works is that these AI models don't actually need to go through clinical trials to get approved. The bar that these models need to clear is not as high as like a pharmaceutical drug, in part because it's not something that you're taking. The risks aren't quite as high as a drug. So like you don't need those clinical trials. But a lot of radiologists want them. They want data that shows without a doubt that these models, if you use them, can sort of reduce advanced disease or death. And part of the issue with that, some of the companies and researchers said, is that these trials do take a very long time. And by the time you have an AI model that you're testing, that model is probably going to be outdated in a couple of years by the time the trial is over. Something else I'll point out is just that the risk based calculators that we talked about earlier also don't have clinical trials done for them. So some of the AI companies, their response to that is why would we be held to a higher standard that these quiz calculators are not?

[Katie Dayton]

How far away are we from seeing these AI screenings actually on the market?

[Brianna Abbott]

So for breast cancer, we're actually probably closer than a lot of people think. Earlier this year, one of the companies, Clarity AI, received FDA authorization to sort of use this model in the clinic. And that was the first one that's gotten approval and they're hoping to introduce it into clinics by the end of the year. And there are a couple of them that are close behind them. So you could be seeing something like this at your routine mammogram somewhere sooner than you think. And then there are some folks that are working on other things, like for lung cancer or even using AI to predict whether or not a person's genetic mutation will actually result in a disease. And those are probably further away.

[Peter Ciampelli]

That was Wall Street Journal reporter Brianna Abbott speaking with our Katie Dayton. And that's it for Tech News Briefing. Today's show was produced by Julie Chang. I'm your host. Peter Ciampelli. Jessica Fenton and Michael Lavalle wrote our theme music. Our development producer is Aisha Al Muslim. Chris Zinsley is the deputy editor and Falana Patterson is the Wall Street Journal's head of news audio. We'll be back later this morning with TNB Tech Minute. Thanks for listening.

[Veeam/GE Vernova Advertiser]

Let's be honest, most HR platforms aren't exactly a joy to use. Deal's different. It's AI native, keeps you compliant and grows with your team, whether you're five people or 50,000. HR IT and payroll on one platform that just works. See for yourself at D E E l dot com.