AI and medicine
equivalent of consultant) radiologist at 98.1%. 3 However, one of the most promising results the investigation produced was a 25.1% drop in false positives, which would undoubtedly improve patient care. This is important, as the UK conducts 2 million mammograms annually; a semi-automated system could massively increase efficiency for both patients and physicians. I think that more frequent mammography check-ups could be conducted on those most at risk to breast cancer (e.g. for genetic reasons) than is currently offered, as an automated AI could rapidly sift through screenings at a greatly increased speed compared to clinicians. This could increase the rates of early detection of breast cancer, saving lives.
An example of AI already in use is OsteoDetect, a program created by Imagen Software that is able to analyse 2D Radiographs (X-Rays) of the wrist, using an algorithm trained by machine learning. It is able to pick up common fractures such as the distal-radium fracture, one of the most common fractures in adults. The US Food and Drug Administration (FDA), which is the federal board that oversees the safety and effectiveness of food and medicines in the United States, gave OsteoDetect the go-ahead for use in clinical environments across the country in 2018, though it is currently used as a purely advisory tool and cannot be used to replace or render obsolete a radiologist analysis of a given radiography. This was after the company provided research detailing the performance of the OsteoDetect algorithm against 3 board-certified hand surgeons in examining 1000 radiographs, and successfully assessing both the presence of a fracture and OsteoDetect’s accuracy of the locali zation of the
fracture. Additionally, the FDA was provided with a retrospective study that summarized the heightened performance of doctors assisted by OsteoDetect in both specificity of the diagnosis as well as sensitivity. 4 Although OsteoDetect has its shortcomings, in that it cannot be used autonomously and must be used as an adjunct tool by a physician within a clinical environment, it shows that there is a growing market for software using machine learning, and that in the coming years the programmes may develop to a point of sophistication that it can be used to determine diagnosis independently of, and without the need of verification from, a healthcare professional. Moreover, it is a piece of AI already being used to great success and provides evidence that AI is effective at increasing accuracy and speed of diagnosis while simultaneously easing the burden on healthcare professionals. Another use of AI in medicine that has been flying under the radar, and that some may use every day without even noticing, is the Apple Watch. If you own one of these, you could receive a notification stating that your heart rate is unusually high or that it has detected an abnormal rhythm. This is due to Apple installing an algorithm in its heart rate detector inside the watch which monitors for any cardiac irregularities. This shows the extent to which AI has already developed, to the point that it can be used to identify Atrial-Fibrillation, a common heart disorder which can often be underlying and 3 See Lauritzen, A et al. https://pubs.rsna.org/doi/10.1148/radiol.210948#:~:text=Artificial%20intelligence%20(AI) – based,performed%20consistently%20across%20breast%20densities. 4 See https://www.fda.gov/news-events/press-announcements/fda-permits-marketing-artificial-intelligence- algorithm-aiding-providers-detecting-wrist-fractures.
185
Made with FlippingBook - Online catalogs