machine learning Alphabet's (Google's parent company) health division "Verily" created an eye analyzing machine learning algorithm which can now predict heart disease

[captain_to_fire]

Primary Source: Nature Biomedical Engineering

Secondary Source: 9to5 Google

 

Quote

Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning

 

ABSTRACT

Traditionally, medical discoveries are made by observing associations, making hypotheses from them and then designing and running experiments to test the hypotheses. However, with medical images, observing and quantifying associations can often be difficult because of the wide variety of features, patterns, colours, values and shapes that are present in real dat  Here, we show that deep learning can extract new knowledge from retinal fundus images. Using deep-learning models trained on data from 284,335 patients and validated on two independent datasets of 12,026 and 999 patients, we predicted cardiovascular risk factors not previously thought to be present or quantifiable in retinal images, such as age (mean absolute error within 3.26  years), gender (area under the receiver operating characteristic curve (AUC) =  0.97), smoking status (AUC =  0.71), systolic blood pressure (mean absolute error within 11.23 mmHg) and major adverse cardiac events (AUC =  0.70). We also show that  the trained deep-learning models used anatomical features, such as the optic disc or blood vessels, to generate each prediction.

This are the times that I have a profound appreciation of technology especially if it's used to save people's lives just like how an Apple Watch saved a man from heart attack.

But before anyone jumps that this is the pseudomedical practice of "Iridology" where a medical charlatan uses patterns in the eyes to diagnose systemic health problems and later prescribe snake oil supplements, this is not one of those. Rather, Google trained their machine learning algorithms by looking into Opthalmoscopy results of 300,000 people and specifically look at the retinal fundus.

The researchers used the standard model libraries and scripts in Tensor Flow.

Quote

Because retinal images alone were sufficient to predict several cardiovascular risk factors to varying degrees, we reasoned that the images could be correlated directly with cardiovascular events. Therefore, we trained a model to predict the onset of major adverse cardiovascular events (MACE) within five years. This outcome was available only for one of our datasets—the UK Biobank. Because data in the UK Biobank come from a fairly recent study  that recruited relatively healthy participants, MACE were rare (631 events occurred within 5 years of retinal imaging, 150 of which were in the clinical validation dataset).

Our results indicate that the application of deep learning to retinal fundus images alone can be used to predict multiple cardiovascular risk factors, including age, gender and SBP. That these risk factors are core components used in multiple cardiovascular risk calculators indicates that our model can potentially predict cardiovascular risk directly. Despite the promising results, our study has several limitations. First, it only used images with a 45º field of view, and future work could examine the generalizability of these findings to images with  either smaller or larger fields of view. In addition, the overall size of the dataset is relatively small for deep learning. In particular,  although the AUC for cardiovascular events was comparable to that of SCORE, the CIs for both methods were wide. A significantly larger dataset or a population with more cardiovascular events may enable more accurate deep-learning models to be trained and evaluated with high confidence. Another limitation is that some important inputs to existing cardiovascular risk calculators were missing from the datasets.

In summary, we have provided evidence that deep learning may uncover additional signals in retinal images that will allow for better cardiovascular risk stratification. In particular, they could enable  cardiovascular assessment at the population level by leveraging the existing infrastructure used to screen for diabetic eye disease. Our work also suggests avenues of future research into the source of  these associations, and whether they can be used to better understand and prevent cardiovascular disease

It's kinda like when genetic testing center 23andMe offered tests for $100 to look for genetic predispositions of certain diseases like heart disease and cancer until the US FDA shut it down because people are freaking out and making medical decisions based on that which actually makes sense because just because someone with a genetic predisposition to a certain illness doesn't mean they'll have it. There are women with BRCA1 mutation and yet they live their lives normally and don't get a cancer diagnosis. With this one, it looks at pictures inside the eye and they achieved 70% accuracy compared that to a standard basic metabolic panel which achieves 72%.

 

Once Google perfected this, it can become a standard tool for diagnosis and prevention. While most overweight and obese individuals are still at a higher risk for Type 2 Diabetes and Hypertension, there are now people suffering from heart problems despite presenting no physical risk factors. One advantage of this is that it's less expensive and it's not invasive. You don't need to be exposed to radiation for a coronary calcium CT scan to detect calcified plaques (atherosclerosis) within the coronary arteries. But this is not yet final as it need to be compared against existing diagnostic procedures like a 2D Echocardiogram which can detect structural abnormalities in the heart or an angiogram which can diagnose diseases related to blood vessels. But since it can also predict diabetic outcomes, it may as well serve as a diagnostic tool to replace the cumbersome fasting requirements to test glucose tolerance or treat metabolic syndrome earlier before complications get worse.

 

[paddy-stone]

That's pretty damn cool.

[8uhbbhu8]

Damn.... all this cool medical tech starting to come out now.

[suicidalfranco]

and now we wait for apple to optimize this

[captain_to_fire]

3 minutes ago, suicidalfranco said:

and now we wait for apple to optimize this

They probably can with their CoreML API. The problem is that at the moment only the iMac Pro can do proper machine learning computations. 

[paddy-stone]

20 minutes ago, 8uhbbhu8 said:

Damn.... all this cool medical tech starting to come out now.

Indeed, I am most impressed by some dental tech, where they found a way to get your teeth to repair holes and such... It'll probably be a few years yet before we actually see it, but that would be so cool. Imagine, going to your dentist and you have a cavity, he drills out the cavity and then injects/puts some stuff on the tooth to make it repair itself.. I would pay for that any day rather than having emalgam fillings. I would imagine they would probably have to put a thin covering "cap" on it while it's repairing.

I really hope that tech come out, it won't put dentists out of business, especially here in the UK where there aren't even enough dentists to go around, and huge waiting times to get registered and such. I have had to go private for my dental needs, as I couldn't get my teeth fixed if not, and it's bloody expensive. I'm glad I don't have a lot of dental problems and looked after my teeth pretty well. I would still welcome the above tech and have my old fillings taken out and have the tooth repaired instead.

[iamdarkyoshi]

25 minutes ago, suicidalfranco said:

and now we wait for apple to optimize this

Until it scans an indian's eye and crashes /s

 

Knowing google, the project will go unfinished until someone else takes ahold of it...

[Doobeedoo]

Yeah much can be learned from persons eye about their health. Cool to see machine learning as well. I mean it will be applies everywhere eventually. 

[BuckGup]

Didn't know you could tell gender and BMI from your eye

[captain_to_fire]

1 minute ago, BuckGup said:

Didn't know you could tell gender and BMI from your eye

I can already see certain groups of people protesting against Google

[BuckGup]

Just now, hey_yo_ said:

I can already see certain groups of people protesting against Google

Lol right let me rephrase that. You can tell sex by looking at your eye because sex and gender are different things now days

[captain_to_fire]

Just now, BuckGup said:

Lol right let me rephrase that. You can tell sex by looking at your eye because sex and gender are different things now days

"Did you just assumed my gender you cis-gender white man?" /s

[Nicnac]

There is so much you can do with ML in the medical field rn.... wish I was able to program tensorflow and be done with my degree already......

[captain_to_fire]

Just now, Nicnac said:

There is so much you can do with ML in the medical field rn.... wish I was able to program tensorflow and be done with my degree already......

ML can be used for vaccine and drug design where pharma companies can use ML models to predict genetic variability of the flu virus so that flu shots have a higher efficacy and would rely less on chicken eggs. It makes me somehow regret taking Biochemistry as a college degree and I probably should have picked something related to programming and I'll probably earn more.

[Jtalk4456]

this is just amazing

[Nicnac]

5 minutes ago, hey_yo_ said:

ML can be used for vaccine and drug design where pharma companies can use ML models to predict genetic variability of the flu virus so that flu shots have a higher efficacy and would rely less on chicken eggs. It makes me somehow regret taking Biochemistry as a college degree and I probably should have picked something related to programming and I'll probably earn more.

yea better flu vaccines would be great! 30% efficacy aren't especially encouraging to go through the extra effort, although one of our professors always emphasises how important and useful they are....

But hey, it's never too late to start, especially programming is easily self-taught