“Machine learning” is a term for a type of artificial intelligence or “AI” that adapts and changes how it behaves over time.
Usually, programs run in predictable ways, the same way every time, depending on what a human operator puts into them. Machine-learning allows the program to learn from its experiences and change over time, fine tuning itself into dealing with situations more efficiently and effectively.
There are many potential applications, even in cooking. But for people with SLE (systemic lupus erythematosus), machine learning may help unravel the mysteries of the disease:
- More accurately diagnose lupus and other autoimmune diseases
- Identify effective and personalized treatments
- Predict changes in disease activity and predict flares
- You can read more about some advanced flare prediction tests here.
- Determine lifestyle programs that provide the most benefit
- Identify triggers
Machine Learning and Lupus
The idea of machine learning was first conceptualized by Arthur Samuel in 1959, and refers to the ability of a machine to “think.” This thinking is the ability of an artificial intelligence program to analyze large amounts of complex data and adapt its programming to fit the data. This type of analysis of large, complex data sets is common in healthcare.
Machine learning allows a program to not only recognize patterns and trends and predict the results accurately, but to improve their accuracy over time as it “remembers” previous predictions. This could revolutionize medicine, which relies almost completely on trained human operators and is subject to human error.
Artificial intelligence and machine learning has the potential to make hospitals more efficient, more effective, and help more people. For some AIs, the goal is to have them act as part of the triage model, making sure that patients get the right diagnosis and are sent to the right doctors even before they speak to a nurse.
AI is very good at detecting and picking out patterns. This makes AI potentially very useful for classifying disease types, better than humans and current tests alone.
AI can also compare test results and behavior patterns quickly and more accurately than a human on their own, making the program better able to monitor changes in a person’s disease and condition. From this information, a treatment plan can be modified, adjusted, or – if it is working – remain the same, with a lot less guesswork.
AI can use various types of data for evaluation including:
- Genetic informaation
- Social media activity
- Lab tests and bloodwork
- Individual hospital readmission rates
AI can also help humans overcome assumptions, which plagues many people with lupus over their treatment journey.
“Horses, not Zebras.”
Doctors are taught to “listen for horses, not zebras.” That means to assume that the symptoms point to a common disease, not a rare one. Unfortunately, lupus is a “zebra” and is often misdiagnosed and ignored for years. To a machine, however, there are no horses or zebras – only information. An AI might be more likely to detect the constellation of signs and symptoms that hint at lupus – and less likely to ignore symptoms or make assumptions.
In a July 2020 report, particular value was placed on AI’s ability to diagnose and treat kidney-related symptoms of lupus, which are some of the most severe and life-threatening symptoms.
Machine Learning Challenges
According to the scientific publication Nature, one of the big questions surrounding AI is “How will we regulate it?” The other question is “How will we test it?”
The U.S. Food and Drug administration (FDA) and other organizations that regulate medical practices and treatments have begun creating new regulations and processes for these tools. These new regulations complement their policies on medications, processed foods, medical implants, treatments, and additives. According to the FDA, AIs are classified as medical devices. You can read the FDA’s statement on machine learning here.
Regulation is vital in ensuring that science and data drive the development of these new health products. Positive, provable results are key.
Safety is also a concern, both for the patients involved and for the program itself.
Standards of security need to be set, to protect these AI programs (which come into contact with sensitive information) from hackers. Restrictions also need to be in place to ensure that, as the AI programs learn how to work more efficiently, that they grow into programs that are better for the patient. In particular, regulators are concerned about the autonomous devices, that operate with little-to-no human intervention. Finally, how will one pay for it? What will one pay for it? At the same time, the AI technology needs space to grow and evolve and become more effective.
Those standards need to be set, and the field of AI is still very much in creation.
All of these challenges are shared with other new medical technologies. And the balancing act between usefulness and safety can be very difficult. However, one challenge, known as “biocreep,” is unique to AI.
What is Biocreep?
“Biocreep” is when a treatment becomes, over time, worse at predicting or treating disease. All prediction techniques make errors, of course, but AIs used in medicine should have a limited rate of errors. Biocreep can be prevented with special algorithms and restrictions on the program that help guide the AI towards better results.
A Lupus Warrior’s Takeaway
Artificial Intelligence and machine learning is still a young field. There is a long way to go before it becomes a staple of medical practice. However, some limited AI is in active use, helping you navigate your treatment through online scheduling, online check ins, digitization, reminders for follow up appointments and immunization dates, and even determining the dose of a medication you take. While this is not at the level of machine learning, we are likely to see more adaptive AI in medicine in the future, to help take some of the workload off of the shoulders of medical professionals.
However, diagnosis should remain a team effort between a professional treatment team, a patient, and AI algorithms. Over-reliance on AIs can be dangerous, but machine learning reduces the mistakes that that will be made. In the near future the lupus diagnosis may be completely revolutionized. The goal of machine learning is not to replace doctors, but to enhance a lupus treatment team’s ability to take care of a patient.
Retinopathy, damage to the retina, is one of the key early potential...
Infections from pathogens like bacteria, viruses, and fungi are a major issue...
What's that? Hearing loss and other aural phenomenon may impact around 30%...