Artificial intelligence is a hot topic in almost every industry right now, and healthcare is no exception.
Copyright: forbes.com
The big data revolution has transformed manufacturing supply chains, retail advertising and customer service. However, transforming healthcare with AI is a very different and exponentially more difficult challenge. In this article, I’ll explain a few reasons why AI in healthcare poses a steeper climb, as well as the potential opportunities that make it worth working toward.
AI In Healthcare: The Challenges
Designing and implementing AI tools in healthcare is fundamentally different from using machine learning or big data in other industries. Let’s examine these challenges using arguably the most robust AI tool on the planet as a counterexample: Google AdWords.
When you search for something on Google and click on a result, every step of that process is captured as data. Every step is also identical for every user — data is captured in completely comparable metrics like clicks.
In healthcare, by contrast, no two patient experiences are alike. Even if you wanted to look only at a standardized doctor-patient interaction, like an annual wellness exam, if the same patient visited two different doctors, each doctor would likely record different information. Even within one provider’s practice, the amount and type of data collected for each patient varies by clinician. Plus, most of the data that affects your health — what you ate, how much you slept, your stress levels, etc. — can’t be collected in a doctor visit. That kind of data, if present in your record at all, is usually based on your imperfect recall.
Not only is the nature of healthcare data more complex and variable, but ethical challenges exist, too. If a Google engineer tweaks the AdWords code and accidentally makes something about the tool worse, they might get fired, and the company might lose money until the problem is fixed. But the stakes are far higher in healthcare. Problems are literally life or death. Mistakes are far costlier than missed revenue. Healthcare data is also extremely sensitive. Even experimenting with building a new tool in healthcare requires a great deal of security to protect patient data.
Datasets in healthcare are also inevitably far smaller — and smaller datasets mean less training data for AI algorithms to learn from. Tools like AdWords can compare millions of people’s results, but a clinical trial might involve 50 patients; a massive trial might enroll 1,000 patients. Real-world healthcare datasets do exist with millions of patients, but by the time you filter down to relevant patients for training a particular algorithm, those datasets get reduced to hundreds of thousands or even less. This is many orders of magnitude smaller than what the most popular AI algorithms are trained on in other spaces. Optimization goals are also different in the healthcare context — with a tool like AdWords, accuracy is the highest goal, but in healthcare, while accuracy is important, health outcomes are far more important. In a clinical setting, false negatives carry worse consequences than false positives.
Thank you for reading this post, don't forget to subscribe to our AI NAVIGATOR!
If creating useful AI tools for healthcare is so difficult, why do it? Because the promise is enormous. Right now, the way we practice healthcare, there is almost no systematic way to learn from the vast amounts of data on patient experiences and outcomes that is generated daily. Doctors learn from their personal experience, but at most, a doctor can see a few thousand patients in their career. An AI tool could analyze and learn from the experience of millions of patients and the knowledge of thousands of clinicians, dramatically improving diagnosis and treatment. […]
Read more: www.forbes.com
Artificial intelligence is a hot topic in almost every industry right now, and healthcare is no exception.
Copyright: forbes.com
The big data revolution has transformed manufacturing supply chains, retail advertising and customer service. However, transforming healthcare with AI is a very different and exponentially more difficult challenge. In this article, I’ll explain a few reasons why AI in healthcare poses a steeper climb, as well as the potential opportunities that make it worth working toward.
AI In Healthcare: The Challenges
Designing and implementing AI tools in healthcare is fundamentally different from using machine learning or big data in other industries. Let’s examine these challenges using arguably the most robust AI tool on the planet as a counterexample: Google AdWords.
When you search for something on Google and click on a result, every step of that process is captured as data. Every step is also identical for every user — data is captured in completely comparable metrics like clicks.
In healthcare, by contrast, no two patient experiences are alike. Even if you wanted to look only at a standardized doctor-patient interaction, like an annual wellness exam, if the same patient visited two different doctors, each doctor would likely record different information. Even within one provider’s practice, the amount and type of data collected for each patient varies by clinician. Plus, most of the data that affects your health — what you ate, how much you slept, your stress levels, etc. — can’t be collected in a doctor visit. That kind of data, if present in your record at all, is usually based on your imperfect recall.
Not only is the nature of healthcare data more complex and variable, but ethical challenges exist, too. If a Google engineer tweaks the AdWords code and accidentally makes something about the tool worse, they might get fired, and the company might lose money until the problem is fixed. But the stakes are far higher in healthcare. Problems are literally life or death. Mistakes are far costlier than missed revenue. Healthcare data is also extremely sensitive. Even experimenting with building a new tool in healthcare requires a great deal of security to protect patient data.
Datasets in healthcare are also inevitably far smaller — and smaller datasets mean less training data for AI algorithms to learn from. Tools like AdWords can compare millions of people’s results, but a clinical trial might involve 50 patients; a massive trial might enroll 1,000 patients. Real-world healthcare datasets do exist with millions of patients, but by the time you filter down to relevant patients for training a particular algorithm, those datasets get reduced to hundreds of thousands or even less. This is many orders of magnitude smaller than what the most popular AI algorithms are trained on in other spaces. Optimization goals are also different in the healthcare context — with a tool like AdWords, accuracy is the highest goal, but in healthcare, while accuracy is important, health outcomes are far more important. In a clinical setting, false negatives carry worse consequences than false positives.
Thank you for reading this post, don't forget to subscribe to our AI NAVIGATOR!
AI In Healthcare: The Opportunities
If creating useful AI tools for healthcare is so difficult, why do it? Because the promise is enormous. Right now, the way we practice healthcare, there is almost no systematic way to learn from the vast amounts of data on patient experiences and outcomes that is generated daily. Doctors learn from their personal experience, but at most, a doctor can see a few thousand patients in their career. An AI tool could analyze and learn from the experience of millions of patients and the knowledge of thousands of clinicians, dramatically improving diagnosis and treatment. […]
Read more: www.forbes.com
Share this: