The marriage between deep-learning artificial intelligence and medicine has always seemed inevitable. Deep learning algorithms could potentially reduce the number of other healthcare resources used. Likewise, these systems could improve efficiency while freeing up time for doctors to be with patients. What’s not to get excited about? But many questions remain regarding the quality of deep learning algorithms in medicine. Thus, it is not yet clear if AI doctors will truly provide obvious benefits over existing healthcare systems.
A major part of the problem has been the quality of the existing studies investigating deep learning algorithms in healthcare. Lack of controls, inconsistent protocols, and a number of other problems plague much of the existing studies. However, research reviews of the few qualitative studies available are beginning to suggest that AI doctor advantages might exist. With this in mind, examining the current state of affairs and potential future developments offers some additional insights.
What Are Deep Learning Algorithms and AI?
Artificial intelligence has been a buzz word in numerous industries for many years. But deep learning algorithms really came on the scene in 2012. In essence, deep learning algorithms represent an AI technique where focused pattern recognition can occur. As a result, many industries, including healthcare, have been excited about the potential of deep learning algorithms. The potential for these systems to develop into an AI doctor type of program naturally has some appeal.
While advances and AI doctors sound intriguing, deep learning algorithms are not the same as human diagnostic strategies. In humans, a cause-and-effect process is inherently present that allows physicians to link findings, processes, and outcomes. In contrast, deep learning algorithms simply use pattern recognition alone. For example, humans can appreciate that dropping a glass will cause it to break after one attempt. Deep learning algorithms would have to repeatedly perform the experiment before the recognition of a pattern. This reflects a major limitation and difference between human doctors and a potential AI doctor.
Recent Research Investigating AI Doctors
Recent research evaluating whether AI doctors can compete with human ones reviews several deep learning algorithms studies. Between 2012 and 2019, fourteen quality studies involving AI imaging were identified. When analyzed, it showed AI doctors using deep learning algorithms correctly identified disease on diagnostic images 87% of the time. Their human counterparts scored 86% in this regard. Thus, it would appear that AI doctors and human ones are comparable in detecting disease on diagnostic images.
While this seems incredible at first glance, some notable caveats must be mentioned. In these studies, both the AI doctors and human physicians used only pattern recognition to detect potential disease. In other words, the normal patient history and additional pieces of information doctors normally have were not available. It would have significantly increased the performance of human physicians. Likewise, AI systems are unable to explain why an image finding causes a patient’s symptom or disease. These are notable limitations that keep AI doctors on the sideline for now.
Other Potential Benefits for Deep Learning Algorithms
While a deep learning algorithm’s diagnosis of patients currently has notable limitations, other potential advantages do exist. Specifically, the use of AI systems includes detecting rare disease states and to help develop pharmaceuticals for them. Bayer is actively using deep learning algorithms for this purpose. Likewise, nearly 150 AI startup companies are currently involved in drug development. In essence, deep learning algorithms are better able to recognize aberrant patterns associated with disease states at cellular levels. This data is then used to help make target drug development more efficient, more precise, and less costly.
While several AI startups are making headlines in this area, two specific ones have gained recent attention. Atomwise is an AI startup that recently partnered with Jiangsu Hansoh Pharmaceutical Group. The $1.5 billion joint venture is invested in developing new cancer drugs using deep learning algorithms. Likewise, Deep Genomics has been using deep learning algorithms for drug development for the past five years. It is currently making tremendous headway in developing a drug for Wilson’s Disease, a rare metabolic disorder. Both highlight how deep learning algorithms are advancing healthcare beyond AI doctor considerations.
AI Doctors Will Have to Wait…For Now
Without question, deep learning algorithms and AI offer great promise in several areas of healthcare. Enhanced pattern recognition can aid doctors in detecting disease and abnormalities. But it is questionable if this is even cost-effective. Such technologies are expensive, and the additional yield appears to be small. Likewise, rural areas with limited providers might be better served through telehealth as opposed to these more costly services. For now, it looks like research and drug development are the most promising fields in healthcare for this technology.
Of course, that could rapidly change if deep learning systems improve. If they advance from pattern recognition to a cause-and-effect model, reassessments will be required. For now, there is a need to develop such abilities, and therefore, artificial intelligence doctors offer little advantage over human physicians. But researchers are actively trying to advance these systems so that such capacities become a reality. While the possibility seems intriguing, the chance artificial intelligence doctors will replace real ones appears to be well into the future. At best, they will likely provide an additional technology to enhance the quality of care. And in the process, hopefully reduce the excessive costs of the entire healthcare system.
