Not for Primary Care - The Promise of AI.

By Marc Torok,

Healthcare AI specialist.

Feb 13^th, 2019.

The RSNA 2018 Annual Meeting, demonstrated the rapid rise of AI and ML in Healthcare. Around the world, the application of artificial intelligence (AI) and machine learning (ML) in medical imaging is a hot topic. While research work will undoubtedly continue and expand, there are also growing examples of AI and ML being used in real-world, clinical settings. It also targets where it is not being applied – Primary Care!

Thus far, two main areas of focus have emerged as the most promising for AI and ML applicability, the first is healthcare information and second is medical imaging. Healthcare produces a wealth of disconnected patient-related information where AI and ML promise to find trends and patterns that could lead to better patient management and better clinical decisions. Whereas in medical imaging, AI and ML are leading to better detection of disease conditions in diagnostic images that promise to assist the specialists.

The progress has been rapid, as the goliaths of tech including Googles, Facebook, Amazon, Microsoft, IBM, … and a host of others, rush-in to capture the pole position in the new race into an emerging branch of value healthcare. As AI heads into clinical use, these early use cases are helping both technology vendors and healthcare providers understand how to integrate these systems into its clinical workflows. So too, healthcare institutions are learning how to value and monetize their data, the life-blood of AI in imaging and decision support systems.

And thus far, the promise of AI is all but absent from Primary Care.

The problem - the lack of first-use medical devices designed for use in Primary Care that can be AI augmented, or AI Powered. And the data needed to feed the AI beast. Only two current Primary Care devices are ripe for AI augmentation, the ECG and the stethoscope. Other heart disease-related devices, are not for use in Primary Care, such as the Echocardiograph, CT, MRI, which requires specialized knowledge and training.

The promise of AI in Primary Care should be to increase the devices effectiveness. For ECG this means AI would have to increase specificity, which, at roughly +97%, is already at the top of its game. AI can add little to this. However, ECG is essentially a bio-signal of an electrical nature and therefore limited to diagnosing diseases that only have or leave an electrical trace. ECG cannot identify with the morphological or acoustic aspects of the heart unless the ECG pathways are altered by a disease.

The other measure of ECG is based on sensitivity. Here AI has a modest possibility to increase sensitivity, which typically is between 60-70%, dependent on the device but yet again is limited to the diseases the ECG bio-signal can detect. However, ECG has already been algorithmically enhanced, which means that again AI would unlikely be able to deliver on its promise.

However, ECG’s greatest limitation is that it can only detect about 44% of all common heart diseases typically found in Primary Care. This is understandable as ECG bio-signal is limited by its electrical nature. This also disqualifies it for use as a first-use device for detecting heart disease in general, particularly in Primary Care. It’s simple, with ECG, too many diseases would remain hidden, undetectable. So although AI could make it more sensitive, it does not make more sensitive to a broader and wider range of common heart diseases typically found in Primary Care. AI would fail on its promise.

So current algorithmically augmented ECG devices, like the Glasgow Algorithm produced by the University of Glasgow, although not true AI, already do for ECG what AI promises.   but its effectiveness remains limited to the nature of the device. However, what it can detect, it detects very well – but this is not enough for use as a widespread screening device in Primary Care!

What is needed in Primary Care is an effective first-use device that can detect over 90% of all common heart diseases. 90% is specified because above 90%, the heart diseases become very rare, and very numerous, with very low prevalence in the population. At 90% effectiveness, such a device can be relied upon by Primary Care providers to enable a reliable systematic widespread screening of the at-risk patient population, both symptomatic and asymptomatic. However, even with AI or algorithmically augmented, ECG bio-signals are simply not enough as both the acoustic and morphological aspects of the heart cannot be diagnosed, leaving more than ½ of all common heart diseases undetectable in Primary Care.

New AI powered heart bio-signal enabled first-use devices are needed. Only then, will AI be able to assist Primary Care physicians with their diagnostic responsibilities as it is in Primary Care, where patients have the first contact with the healthcare system, and expect the detection of heart disease onset to be diagnosed first. The promise of AI, is to not betray that trust.