Credit: AI Trends
By Andreas Scherer, Ph.D. President & CEO of Golden Helix
Few topics have captured the imagination of the general public and executives in the healthcare IT world like Artificial Intelligence (AI). Are we at the cusp of creating intelligent systems that reach or surpass the ability of human experts? Are we rendering human decision making obsolete by replacing health care professionals with well-designed and trained AI bots that can tireless work around the clock? Well, it is safe to say that this scenario won’t be reality anytime soon. So, what is all the buzz about this?
Let’s look beyond the Sci-Fi of AI and instead look at Artificial Intelligence from a purely computer science perspective. Essentially, we are talking about a set of computing technologies that are subsumed under the umbrella term machine learning. We distinguish between supervised and unsupervised machine learning as the main categories. In supervised learning, there is a set of input and output data. Think, for example, of a facial recognition system. In this case, the input is a set images of people’s faces and the output is each person’s identity. In supervised learning, the system learns the mapping between input and output data. There are very powerful technologies such as neural networks that can efficiently learn such mappings and are, in addition, able to deal with noisy input data. So, in our face recognition example, it wouldn’t matter that a person is suddenly wearing glasses or is sporting a three-day beard.
In unsupervised learning, we only have input data. The algorithm is given the task to find interesting and relevant structures in the data guided only by some performance metric. One highly publicized AI breakthrough was Google’s AlphaGO AI that taught itself the very complex board game GO by playing against itself. Ultimately, it was able to beat the reigning world champion decisively. A breakthrough was accomplished by employing strategies that had not yet been discovered by humans.
These powerful technologies have a wide application spectrum in the Health Care space. Here are a few examples:
- Robotic surgery: Supporting surgeons in applying robotic surgery techniques by providing better control and optimizing patient outcome by analyzing data from previously conducted surgeries as part of an ongoing learning effort.
- Nursing assistance: Monitoring vital patient data and providing the ability to detect patterns early that potentially warrant a medical intervention
- Medical imaging: Radiology and pathology are being disrupted by accurate labeling of 2D and 3D images and scans by highly trained algorithms. This improves efficiency and broadens access to underserved rural regions.
- Diagnostics and predictive analytics: Medical decisions that require complex analysis have a significant opportunity to reach their full potential. For example, prediction systems can be built for common diseases based on personal and environmental factors.
A fantastic example of the use case of utilizing Artificial Intelligence in the HealthCare sector is in the emerging field of precision medicine. This is a fast-growing clinical segment that is highly data driven. As next-gen sequencing becomes more affordable and the amount of knowledge in this field has exponentially increased since the first human genome was first sequenced in 2003, it is today part of clinical best practice to sift through billions of base pairs and millions of mutations per sample. This is only possible today by leveraging advanced analytics products. As payers, such as Medicaid, are considering the use of next-gen sequencing as part of standard care, the number of samples that are being processed is increasing rapidly. At the same time, the number of experts in this field is only growing modestly if at all. The only way out of this conundrum is to accelerate automation in this field. Since the entire diagnostic process is completely data driven, this field is a prime target for utilizing AI technologies.
In doing so, we must be careful in our implementation approach. Here are some key considerations in taking a cautious approach:
- Human Supervision: Particularly for an emerging fields such as precision medicine, it is crucial that however advanced the system, medical experts have final control over the diagnosis. Even in the most data-driven application areas, AI technologies will not replace diagnostic decisions making by humans any time soon.
- Reproducibility: AI-based systems must be built such that any output can be verified as the diagnostic process unfolds. It’s is crucial that the systems can be verified at any evolutionary stage. As in standard software development, there is a need for versioning of AI components that can be thoroughly vetted at each step before being put into production. Ultimately, AI systems will need be introduced into the clinical diagnostic process like any other software tool following current applicable guidelines such as CLIA.
- Data Privacy: We have to put careful safe guards in place that don’t allow de-anonymization of personalized information. Some AI technologies have very powerful obfuscation capabilities that could be used for this purpose.
The usage of Artificial Intelligence in the healthcare sector is inescapable. There is a plethora of application areas where AI-based systems help to increase throughput, help to eliminate or at least reduce human error and make expertise available that is scarce or entirely unavailable in particular geographic locations. The burden falls on technology companies to leverage these technologies with care. Hospitals and labs, as the implementers of this paradigm, are well-advised to introduce changes within the current guidelines while ensuring the appropriate checks and balances are in place. Finally, law makers need to stay ahead of the technology curve to create an adequate legislative and regulatory framework as the usage of AI technology will grow over the years to come.
Andreas Scherer, Ph.D. is the President & CEO of Golden Helix. In the last decade, he has focused on accelerating R&D processes of Fortune 500 pharmaceutical, biotech, and medical device companies.
See the source post at Bio-IT World.