In an article recently written by Olivier Salvado, Group Leader for Biomedical Informatics (CSIRO), for The Conversation, the author explains why artificial intelligence has not revolutionised healthcare yet. Salvado highlights the various areas that appears to be challenging in today’s healthcare-system, but also explaining the measures taken to address these issues, and where the current technology is at the moment.
With the introduction of deep learning the limits of what machines can do has been pushed, as the technology comprise neuron-like units into hierarchical layers which then can recognise patterns in data.
“This is done by iteratively presenting data along with the correct answer to the network until its internal parameters, the weights linking the artificial neurons, are optimised”, Salvado writes.
“If the training data is able to capture the variability of the real-world, the network is able to generalise well and provide the correct answer when presented with unseen data”.
This means that the learning stage is depending on receiving very large data sets along with corresponding answers, to work properly. Millions of records and billions of computations are required to update the network parameters, which is often conducted on a “supercomputer”, for days or weeks.
This is where the problem with healthcare is, that the data sets are simply not large enough yet, and the correct answers to be learned are often ambiguous or in some cases, unknown.
Measures are taken to increase the data sets in medical care, with initiatives such as Biobank in the United Kingdom, which aims to scan 100,000 participants. Others include the Alzheimer’s Disease Neuroimaging Initiative (ADNI) in the United States and the Australian Imaging, Biomarkers and Lifestyle Study of Ageing (AIBL), tracking more than a thousand subjects over a decade.
What a machine needs to learn is not obvious:
Salvado uses the example of when radiologists analyse an X-ray in order to reach consensus, to explain how what a machine needs to learn is not obvious. When radiologists disagree when interpreting a scan, due to blurred or subtle features for instance, incidents where the true answer can be obtained, might not be there at all.
An example is when measuring the size of a structure from a brain MRI cannot be validated, Salvado writes, even at autopsy, since post-mortem tissues change in their composition and size after death.
This is where machine learning comes too short, Salvado explains, writing: “It is a much more difficult task to measure the size of a brain structure from an MRI because no one knows the answer and only consensus from several experts can be assembled at best, and at a great cost”.
However, this issue is also being addressed, as several technologies are now emerging in order to solve it. Complex mathematical models including probabilities such as Bayesian approaches can learn under uncertainty.
“Another approach is transfer learning, whereby a machine can learn from large, different, but relevant, data sets for which the training answers are known”, Salvado writes.
“Medical applications of deep learning have already been very successful. They often come first at competitions during scientific meetings where data sets are made available and the evaluation of submitted results revealed during the conference”.
Salvado explains that the most challenging issue of all, is about understanding causation. “Analysing retrospective data is prone to learning spurious correlation and missing the underlying cause for diseases or effect of treatments”, he writes.
Normally, randomised clinical trials will provide the research with evidence of the superiority of various options, but they do not still not benefit from the potential of artificial intelligence, Salvado explains.
However, new designs such as platform clinical trials might address this issue in the future which could potentially pave the way of how machine learning technologies could learn evidence, rather than just association.
“So large medical data sets are being assembled. New technologies to overcome the lack of certainty are being developed. Novel ways to establish causation are emerging”, Salvado finishes.
“This area is moving fast and tremendous potential exists for improving efficiency and health. Indeed many ventures are trying to capitalise on this”.
This article was first published at: http://theconversation.com/why-artificial-intelligence-has-not-revolutionised-healthcare-yet-69403