Laboratory testing is usually regarded as being on the cutting edge. New technologies are constantly emerging in the field, and testing modalities are always improving, but there are still some advances that stand to change the nature of the clinical laboratory. Artificial intelligence (AI) and Machine Learning (ML) are trends that make it seems like science fiction is slowly becoming reality. Not only are these modalities on the horizon, but some of their results are already in use -- allowing for continued growth in the industry.
"The big thing that the whole industry's moving towards is very high throughput systems. All of the different 'omics' technologies will start to become standard routine clinical testing in a few years," said Jimmy Lin, MD, PhD, MH, president of the Rare Genomics Institute, in a recent interview. "Genomics is starting to be implemented on a regular basis, but all of the other high throughput technologies will. So, with this avalanche of data that's going to come, a lot of different AL tools will start to be implemented and start to be more important."
AI is a large field and can be seen in a number of different applications. From breakthroughs in vision and trying to replicate human vision and understanding language and pattern recognition to the ability to analyze and interpret large amounts of information, AI/ML technologies can be seen all over --a notable example, as Lin discussed, is the Siri app on the iPhone. With the development of whole genome and exome sequence analysis and the impossibly large sets of data that are produced in the testing process, the potential uses of AI/ML in clinical laboratories is beneficial in examination and possibly treatment down the road.
"Within biology, one of its most powerful uses is to take genomic data or transplastomic data or these large data sets," continued Lin. "And being able to figure out unique signatures within them that could be used for diagnosis and clinical testing."
One in a Million
An area of clinical science that utilizes AI tools in terms of interpreting large data sets is bioinformatics, which contributes to higher throughput systems, and transcriptomics, a technique that can highlight important gene patterns. Lin also pointed out that, after clinical trials, simplified versions of transcriptomic patterns can then be used in the clinical setting. Robert M. Wenzel, director of London Global Laboratories, brought up another interesting point in the evolution of AI/ML, discussing the different between strong and weak AI technologies. Stronger AI applications refer to technologies that are still in testing phases, while weaker AI refers to the basic computing that is already being used.
"Another important differentiation is strong and weak AI/ML," said Wenzel. "Whilst strong AL is mostly subject to scientific research, weak AL is already widely used. Depending on the definition of weak AI, you can find it in plenty of products ranging from recommendation systems through search engine technology and decision making systems in medical laboratories."
Lin, on the other hand, discussed the idea of "closing the loop," citing research in treatments for diabetes and pancreatic disorders as an area with a lot of potential. As there are technologies that can detect when the body needs insulin and others that can determine and inject the dose as needed, the obvious solution is to combine to the two technologies and "close the loop" by essentially developing an artificial pancreas. According to Lin, however, there are other factors to consider. As technology can malfunction, how can we know that there won't be a miscalculation and the AI technology that delivers the insulin could give too high or too low a dosage? What if it simply shuts down and no dose is given at all? Lin went on to note that, although these malfunctions may only be one-in-a-million, they can't be ignored -- especially in scenarios that could be life or death.
"The caveat is that sometimes, because the information is so complex, it makes it harder for humans to be able to catch errors and to understand what exactly is going on to be able to use it in an appropriate way," explained Lin.
As it currently stands, clinical AI technology is mainly used as an aid to clinicians in analysis, but as Lin noted, the information gathered in certain types of testing, while it can be analyzed by AI recognition software, can become too large for humans to double check efficiently. Wenzel also commented on the applicability of AI/ML as a way to assist researchers, but made the point that patients and physicians are not confident enough in the new technologies to use them without further examination yet -- even despite AI being more accurate than human diagnosis.
"People don't like to get diagnoses or prescriptions via computer," continued Wenzel. "The system should be designed as a helper system for today's doctors, which leads to the next issue: the trust in such systems has not been given yet."
Laboratory testing modalities are always in a state of transition. As clinicians continue to work with improved, high throughput equipment, stronger AI/ML technologies are being incorporated into standard testing practices. With the introduction of more advanced methods, laboratory professionals, clinicians and physicians have to prepare for the future by figuring out the most helpful potential uses for this new technology.
Michael Jones is on staff at ADVANCE.