Machine learning, a key tool in the realm of artificial intelligence (AI) can be applied to data on weather conditions resulting in a forecast of the weather. Similarly, machine learning can be applied to medical data and enable an AI to learn how to diagnose a medical condition. AI systems are learning to diagnose disease across a wide range of medical conditions, and gradually they are becoming as accurate as human doctors.
A good example of AI diagnosing is occurring in London because of a collaboration between researchers from Google’s DeepMind subsidiary, University College London, and Moorfields Private, the private patient division of Moorfields Eye Hospital. The researchers used deep learning, a form of machine learning, to create an AI which can identify more than 50 common eye diseases based on thousands of 3D scans. With a single scan of a patient’s eye, the AI can recommend a specific treatment. While the research is still in the early stage, the results to date are very promising. The Verge, an American technology news and media network, quoted Dr. Pearse Keane, a Consultant Ophthalmologist at Moorfields who was involved in the research, saying,
The number of eye scans we’re performing is growing at a pace much faster than human experts are able to interpret them. There is a risk that this may cause delays in the diagnosis and treatment of sight-threatening diseases. If we can diagnose and treat eye conditions early, it gives us the best chance of saving people’s sight. With further research it could lead to greater consistency and quality of care for patients with eye problems in the future.
The software the researchers developed uses algorithms which can identify common patterns in data from 3D scans of patients’ eyes. The scans are made using a technology called optical coherence tomography (OCT). The researchers submitted data from nearly 15,000 OCT scans from 7,500 patients. In addition to the data from the scans, the researchers included the software diagnoses made by Moorfields doctors. Based on what the AI software learned from the data, it was able to develop a diagnosis without a new scan. The Verge reported the AI’s diagnoses were 94% accurate when compared to the diagnoses made by a panel of eight doctors.
Assuming the AI achieves equal or better accuracy of diagnoses and the software is rolled out to medical providers, there may remain to be issues of concern about the technology. If an AI makes a diagnosis with no human involvement, who is responsible if the diagnosis turns out to be wrong? Doctors are not comfortable with “black box” diagnoses where the basis of a diagnosis is not known.
The researchers at DeepMind and its collaborators are aware of these issues, and they have developed software mitigations. For example, rather than the AI just providing one diagnosis, it can provide several, including information about how the diagnosis was reached and what level of confidence it included. Another ameliorating consideration is the AI can be used for triage. Rather than acting on an AI diagnosis, a provider may look at the diagnoses which are highlighted as the most urgent and consider treatment of those patients with top priority.
Another issue is the ownership of the OCT scan data. Google might argue it owns the data since it expended the effort to store and analyze it. It might argue it has the right to sell the data to other hospitals. Moorfields might argue it owns the data because it came from its patients. Issues such as this are not uncommon in medical research. They are usually resolved with contractual agreements giving the technology company rights to the data for a specific amount of time and then provide royalties to the medical provider.
Regardless of the various issues, it is clear AI may be of great benefit to patients. The Verge reported,
Some 285 million people around the world are estimated to live with a form of sight loss, and eye disease is the biggest cause of this condition. OCT scans are a great tool for spotting eye disease (5.35 million were performed in the United States alone in 2014), but interpreting this data takes time, creating a bottleneck in the diagnostic process. If algorithms can help triage patients by directing doctors to those most in need of care, it could be incredibly beneficial.
As I wrote in Robot Attitude: How Robots and Artificial Intelligence Will Make Our Lives Better , AI will find its way into every aspect of our personal and professional lives. The potential is there for some bad things to happen, and government and scientific experts must be on the lookout and propose appropriate regulations. However, I believe the benefits are huge, and we should learn about and embrace AI.