Artificial intelligence (AI) in some cases performs better at diagnosing certain conditions than doctors, argues Professor Sir Bruce Keogh of the UK National Health Service (NHS). As a result, the NHS wants to allocate more budget to accelerate the application of AI in healthcare, with the organisation seeing particular opportunities in analysing large amounts of information collected on patients' symptoms. What is AI being used for in the medical world? We list some examples.
Sir Keogh makes his statements in an interview with The Guardian. "We know through multiple studies that AI, under certain circumstances, is better at diagnosing certain conditions than doctors," Sir Keogh said. "It is possible that some types - and there are many different types of AI - will be able to interpret X-rays. I understand from people involved in developing this kind of thing that this could be possible in a four-year time frame. We also know that this kind of development can be used for reading histopathological images."
Analysing medical images
A variety of companies and organisations are focusing on developing AI applications that can help improve medical care. One example is Dutch start-up Aidence, which raised €2.25 million from investors as recently as June. This company is working on AI capable of interpreting X-ray, MRI and CT images. With this, Aidence aims to reduce the number of misdiagnoses, increase efficiency and reduce the shortage of radiologists worldwide.
Initially, the company is working to market an AI-based product for detecting lung nodules. However, according to Aidence, the technology used for this can also be used to assess CT scans of the thorax and MRI images of both joints and the brain.
Detecting signs of heart attacks
South Korea's Mediplex Sejong Hospital, in collaboration with startup VUNO Korea, has developed an AI system capable of detecting omens of a heart attack 24 hours v before the attack actually occurs. This allows doctors to take measures in advance of a heart attack.
The system is called cardiac Arrest Early warning alGorithm In Sejong hospital' (AEGIS) and uses machine learning to analyse a patient's breathing, heart rate, oxygen saturation and blood pressure, among others, in real-time. By detecting abnormalities in these, it allows early detection of signals that may indicate a heart attack.
Incidentally, the Modified Early Warning System (MEWS) has long been used to detect omens of heart attacks. In an interview with Korea Biomedical Review, Park Jin-sik, board chairman of Mediplex Sejong Hospital, indicates that heart attacks can be predicted with an accuracy of about 55% using MEWS. AEGIS increases this accuracy to 70%, according to Jin-sik.
Detecting cardiac arrhythmias
Another example is a deep learning algorithm from US Stanford University, which is able to detect 14 types of cardiac arrhythmia. Deep learning is a technique that uses an artificial neural network based on a biological neural network. This network consists of multiple layers of computer systems, each performing a different analysis. Each layer processes the output of the previous layer, constantly refining the analysis. This network can be trained by having it process large amounts of data. The more data the network processes, the smarter it theoretically becomes. This process is called deep learning. Deep learning is a form of machine learning.
An electrocardiogram (ECG) is usually taken from patients suffering from cardiac arrhythmias. If the problem is not visible on this, patients in many cases wear a wearable device that continuously monitors their heart for a fortnight. This generates hundreds of hours of data, making analysis of the collected data time-consuming. In addition, serious problems are not always easy to distinguish from harmless irregularities in the patient's heart rhythm.
Stanford University researchers have succeeded in developing a deep learning algorithm that can analyse this data and recognise 14 different types of cardiac arrhythmia in it. Not only does the algorithm save cardiologists a lot of time, it can detect all 14 cardiac arrhythmias as accurately as cardiologists. Some disorders are even detected more accurately.
Analyse genome data
A glioblastoma multiforme (GBM) is the most aggressive and fastest-growing form of brain tumour. To treat GBMs, genome tests are performed, which identify mutations in the patient's genes. Based on these mutations, doctors can identify medications suitable for treating the specific patient.
However, assessing genome data is very time-consuming, so in practice, genome studies only look at a small proportion of a patient's genes. To improve genome research, researchers at IBM Research are deploying machine learning to analyse genome data. In analysing this data, they used IBM Watson, a supercomputer specifically dedicated to machine learning.
The deployment of IBM Watson allows genome data to be analysed much faster, making it possible to examine more of a patient's genes and get a more complete picture of all the mutations that have occurred in a patient's genes. As a result, more potential drugs suitable for the patient's treatment can be found.
Author: Wouter Hoeffnagel
Source: The Guardian
Source: HenQ
Source: Aidence
Source: Mediplex Sejong Hospital
Source: Korea Biomedical Review
Source: Stanford University
Source: IBM Research
Source: Neurology Genetics
Source photo: Pixabay / GDJ
