'The accuracy of medical diagnosis by AI is only as high as that of a human doctor'
by
Success of artificial intelligence (AI) is being expected to be in the medical field, AI learned based on an enormous amount of image and data is research that's superior to human doctors also report has been. However, research results have been reported that many of the papers researching the application of AI to medical settings are insufficient, and in fact, AI and human diagnostic accuracy do not change .
A comparison of deep learning performance against health-care professionals in detecting diseases from medical imaging: a systematic review and meta-analysis
https://www.thelancet.com/journals/landig/article/PIIS2589-7500(19)30123-2/fulltext
AI equal with human experts in medical diagnosis, study finds | Technology | The Guardian
https://www.theguardian.com/technology/2019/sep/24/ai-equal-with-human-experts-in-medical-diagnosis-study-finds
Research on medical diagnosis by AI has advanced rapidly in recent years, and there are many studies that conclude that “AI can diagnose with better accuracy than humans”. For example, it was reported that AI for detecting lung cancer developed by Google boasts a higher detection rate than human doctors.
Research result that Google's lung cancer detection AI is already superior to human doctors-GIGAZINE
In addition, some researchers argue that AI can diagnose with higher accuracy than humans, and that diagnosis by AI increases the number of patients who do not need unnecessary resection surgery due to a diagnosis error.
A case where half of patients who had surgery were able to avoid unnecessary surgery if analyzed with AI-gigazine
Research teams such as Dr. Xiaoxuan Liu and Dr. Alastair Denniston from University Hospitals Birmingham investigated research papers related to deep learning published since 2012. There seem to be more than 20,000 research papers on diagnosis by AI, but `` Research that prepares high quality data and shows human experts the same images that were analyzed by AI 'It was found that only 14 were found.
The research team collected the most promising results from 14 studies, and then calculated the accuracy with which AI accurately diagnosed human diseases. Then, human doctors had made an accurate examination with an accuracy of 86%, whereas AI was an accuracy of 87%. In addition, the diagnosis accuracy for complete cure of the disease was 91% for humans, but 93% for AI. In other words, when considering only research that is rich in content, there was no significant difference in the accuracy of diagnosis by human doctors and AI. However, the research team added, “Because this result was derived from a small number of studies, we need to be careful.”
Professor David Spiegelhalter at the University of Cambridge points out that research on the application of AI in the medical field is inadequate. “Dr. Liu's research proves that the quality of most evaluation studies is at a sad level through hype for AI in medicine.” “Deep learning may be a promising technology, but clinicians and The commissioner should argue, “What are the benefits of clinical examination with AI?”
by TheDigitalArtist
However, Dr. Denniston has a positive view of the potential of AI in medical care, and says that AI functions as a diagnostic tool and may be useful for efforts such as scanning medical images and backlogging. Dr. Liu also believes that 'if you have AI, you can do it even if you can't have human resources to do diagnostic imaging.'
Dr. Raj Jena, an oncologist at the University of Cambridge, emphasizes that AI needs to be tested to make sure that AI is robust in the future, even though it recognizes that AI will be an important future in the field of medicine. “It is important to understand why AI makes the wrong evaluation,” he said.
Related Posts: