Machine Learning And Healthcare
Industries are changing, thanks to technology, but not everyone is talking about the key values major advents like machine learning in medicine bring to the game. There’s a lot of speculation surrounding this game but a lot of it is nonsense.
The world is moving toward using machine learning (ML) and artificial intelligence (AI) just about everywhere – the application to physical and mental health fields is nothing short of extremely promising. Let’s quickly revisit some of the content we’ve produced on the matter then delve into how these systems will become staples in viable healthcare solutions.
How AI works in the mental and physical healthcare field
Much like a human, an ML or AI system has the capability of learning which is somewhat similar to how an organism learns. For example, when you dedicate a memory, you use one or more of several processes to retain a memory.
However, unlike how information is interpreted by an organism, ML and AI learn from data that’s input where the backend system is responsible for how the data is qualified or quantified. Technically, ML and AI are two different things (though we are using the terms interchangeably in this piece) but the fact remains that in either case, such systems have to be “fed” information for it to learn. This means the backend needs to be accurate in how the system is categorizing information – if you want a system to be able to recognize cars but it’s getting pictures of giraffes in the mix, you’re going to run into issues.
In the human brain, the process isn’t always accurate – mental distress and other issues can obfuscate how memories are retained. Criminal and civil proceedings can call on a child for testimony under the excited utterance clause in many states. This is because we recognize that during ‘startling events’ our mind doesn’t have time to get into what Nobel Prize winner and world-renowned psychologist, Dan Kahneman, describes as “system 2 thinking” which is when thoughts go into literally and figuratively, the more analytical portions of our brain.
Right now, our AI systems can be trained to work through logic and other higher forms of thinking, but the technology is about as evolved as a 9-year-old. Our systems can recognize patterns and correlate information (much like those in what Piaget would describe as the concrete operational stage of cognitive development.) There is a degree of reason that’s involved in decision-making or critical thinking processes but it’s heavily dependent on the data that was used to train the system (i.e. the child’s mind) in the first place.
We have to consider the current limitations of this technology but also recognize the benefits our existing systems provide. From here, we can build better M-based applications.
What is machine learning in medicine accomplishing?
There are a lot of reservations about AI, especially with machine learning in medicine. Many of us have been conditioned by science fiction and worse, Elon Musk’s rant on Joe Rogan’s podcast among other speculations makes the technology feel horrifying.
One of the supposedly smartest men in the world smokes a joint with MMA host/DMT-aficionado, Joe Rogan, and it feels scary. We’re basically in charge of how these systems learn, so the premise that we’re under the control of these systems is mentally troubling.
But if we look at what is actually going on in the machine learning in the medical field (among others) the value these systems can provide. If we pump the brakes on our speculative tendencies, we can see what this tech is actually providing today and ideally, keep everything in check before Terminators come back in time to kill us for some random action we took.
Language processing for expediting diagnoses and treating illness. The example I have here isn’t directly medical, but it’s a significant underlying component of AI and medical machine learning. One way that AI-based systems in security and machine learning in medicine are helping reduce inaccuracies (as well as straight-up fraud) is through a process known as natural language processing.
You’re probably already familiar with this on a consumer basis – software like Google Assistant, Siri, Alexa and so on take natural, spoken commands (both as questions and statements) to perform certain tasks, thanks to natural language processing (NLP.) While these systems are geared toward entertainment and convenience, there is value to these kinds of systems you may not realize.
In the example of the cybersecurity platform, Inky, their system aims to find fraudulent activity before and even after an account is compromised. Say Stephanie in accounting is suddenly using words not typically found in her vernacular, for example, she calls you ‘bro,” says “that’s what’s up”, or uses “lol” when she’s more of a “haha” person, this system will detect such language anomalies and alert the proper people that there is a breach (or compromised account) within the organization.
In the medical field, AI is helping profile those with certain cognitive issues. For example, someone with a dwindling cognitive state due to dementia, Alzheimer’s or another condition can be diagnosed quicker when they seem “off” prior to leaving a stove burner on for days or forgetting to feed the dog.
There are other applications such as with developmental psychology which is especially critical in recognizing those with autism and other learning disabilities. By assessing the manner in which a person communicates, revealing traits become more apparent, granting those in the medical field increased insight into a person’s condition, enabling treatment regimens to be further refined which can provide the individual a better course of treatment.
Computer vision for medical imagery. One of the biggest faults that stem from diagnostic measures is the human element – inexperienced individuals may tend to overfocus on something that surfaces in an Xray, cat scan, MRI, etc. while more jaded professionals tend to overlook certain tell-tale signs of a condition because they’ve “seen it before.”
Medical professionals are tasked with the burden of memorization and pattern recognition. Every individual’s “scan” is a bit different but signs of disease are usually consistent. It’s up to the professionals to disseminate characteristics from an individual’s physical readouts against medical imagery.
Rather than rely on some inexperienced person’s perception or perhaps a jaded medical industry professional, computer vision can assess images from various sources to provide preliminary insight into an individual’s condition. Certain, minute issues surface early with some diseases and they’re simply not immediately obvious to even the most trained eye. Systems that can parse visual information to the pixel can help thwart delayed treatment of potentially life-threatening conditions, giving patients a better quality of life throughout their darkest times.
Safely cross analyzing conditions through patient-to-patient comparison. One of the big issues in the Internet landscape is how our information is collected and used which does mean the possibility of data being leaked, misused or revealed through a data breach. Some outright neglect platforms like Facebook, Twitter, and other social sites because of this fear but in order ML and AI systems to reach full capacity, analyzing large datasets of non-personally identifiable information (non-PII) is necessary as this will serve to help medical professionals diagnose issues as well as uncover comorbid problems faster and with greater accuracy.
Data breaches can happen but this is why we have certain laws and regulations (like HIPPA) in effect to help prevent the disclosure of private, personally identifiable (PII) health information. It’s up to developers, network admins and others in the IT realm to make sure data is as safe as possible.
Most digital platforms often share information, both PII and non-PII, which is incredibly helpful for digital advertisers – this is why you can be scrolling through Facebook and see an item or service surface that you had been searching for or maybe even stumbled upon. Similarly, the non-PII medical information that’s aggregated and collected by medical systems will unveil data to a provider that may have overlooked or undervalued.
For example, a patient may show high blood pressure but this can reflect more than one problem (or something different than an initial diagnosis) as multiple physical issues have similar symptoms, whether cardiovascular disease, glaucoma, general anxiety disorder (GAD), or any other number of problems. You wouldn’t want to be treated for anxiety and prescribed an SSRI when you actually need blood pressure medication!
Patients who are willing to be monitored and have their non-PII information collected and interpreted by non-invasive systems can help to serve others dealing with similar symptoms. In time, machine learning in medicine will help improve the overall quality of physical and mental healthcare.
Learn how Blue Label Labs is using AI and ML to solve healthcare problems
The notion of AI and ML seems like problematic technology when improperly positioned but it truly has the capacity to change the quality of our lives when appropriately used for the better good. The applications that language recognition, computer vision, and wide-scale analytics can provide are immense.
Talk to us at Blue Label Labs to learn about the work we are doing in the healthcare field such as incorporating AI and ML into the applications we build.