Machine learning: what’s the diagnosis?

It also frees up healthcare resources and money if these people never even become patients. Here are some of the illnesses that may become a thing of the past if AI is able to predict them correctly:

Heart attacks can occur with little to no warning and lead to 200 deaths a week in the UK; however, they are largely caused by patient behaviours, such as their diet. While doctors try to predict who is at risk based on health and behaviours, humans only have a 30% success rate. But now, a machine-learning algorithm created by Carnegie Mellon University looks at 72 parameters in patients’ medical history including vital signs, age, blood glucose and platelet counts, and then assesses whether the patient is heading for a ‘Code Blue’ attack. When tested on historic data the system was able to tell, sometimes 4 hours before the event, whether a patient would have gone into arrest at 80% accuracy. When implemented into hospitals, this accuracy and early warning system will be able to save countless lives and shorten hospital stays.

Machine learning isn’t restricted to analysing physical ailments using patient data. In 2015, a team of researchers developed an AI model that correctly predicted which members of a group of young people would develop psychosis—a major feature of schizophrenia—by analysing transcripts of their speech. This model focused on tell-tale verbal tics of psychosis such as short sentences, frequent use of words like “this,” “that,” and “a”, as well as a muddled sense of meaning from one sentence to the next. This then allows for better evidence in diagnoses as mental health issues are often harder to test for, allowing for a speedier path to a treatment stage for patients instead of too much time at diagnosis stage.

Breast cancer is the most common cancer in the UK, with one in eight women receiving the diagnosis in their lifetime. Frequent mammograms are now common in the UK, and so doctors have access to data that shows which women are at most risk. However, reviewing these charts takes a significant amount of a specialist’s time - manual review of 50 charts can take two clinicians 50 to 70 hours. Reviewing the charts isn’t just time intensive – the charts can often be misinterpreted too. In the US, 12.1 million mammograms are performed annually, but half yield false results according to the American Cancer Society, resulting in one in two healthy women being told they have cancer and often having to undergo invasive surgery.

Here too, AI can save time and increase accuracy. Researchers at Houston Methodist Research Institute in Texas have developed artificial intelligence software that reliably interprets mammograms and translates patient data into diagnostic information 30 times faster than a human doctor, with 99 per cent accuracy.  Clinicians use results produced by the AI software, such the expression of tumour proteins, to accurately predict each patient's probability of breast cancer diagnosis.

As well as decreasing the number of false results, the AI software can review 500 charts in a few hours, saving doctors 500 hours of their time.

Treating type 2 diabetes costs the NHS £8.8bn each year, and, despite being largely preventable, its prevalence continues to increase. In Salford, UK, one in ten men over the age of 60 has diabetes. One of the major drivers for the development of diabetes is obesity, which increases the risk of developing the disease. In partnership with Hitachi, Salford Royal Foundation Trust and Salford Clinical Commissioning Group, Salford now has an integrated electronic records system, meaning that people at greatest risk of developing type 2 diabetes can be identified and supported to work towards healthy lifestyle goals. The system leverages patient data, including weight and glucose regulation, to work out whether they are at risk of developing the disease and then makes lifestyle suggestions to become healthier to prevent this.

The use of AI in healthcare has the potential to make the most impact on our day to day lives. By letting patients know what disease they are heading for based on data rather than judgement, they should be more likely to make lifestyle changes and react accordingly. Local authorities can also save money on analysis and diagnosis, allowing resources to be spent looking after those who could not make the changes necessary. An apple a day keeps the doctor away, but a computer might have different ideas about what’s best for you.