AI with “imagination” can help doctors diagnose, especially in complex cases

Doctors are overworked and scarce all over the world, but machine learning can soon help to reduce errors in primary care. AI Symptom Checker is of great value in providing users with medical information and safe triage advice. However, there is nothing to diagnose like a doctor. Unlike doctors, existing symptom checkers provide advice based only on correlation. Correlation is not the cause. Babylon researchers have made it possible to diagnose AI-written test cases using principles of causal reasoning that we first knew.

Researchers used a new approach, called causal machine learning, that is gaining attention in the AI ​​community to act as “imagination.” This made it possible for the AI ​​to consider what the symptoms would be like in the case of a different disease than the patient. Refereed research published in Natural communicationShows that it unravels correlations from causality to make AI much more accurate.

Dr. Jonathan Rickens, the Babylon scientist and lead author, said: “We used a powerful algorithm to take the AI, imagine another reality, and give it the ability to consider if a symptom is present if it is another illness. This makes AI possible. Breaking down potential causes of a patient’s illness and getting scores higher than 70% of doctors in these writings test case“

As Babylon CEO and founder Dr. Ali Pulsa said: “Half of the world has little access to healthcare. We need better response. It’s exciting to see these promising results in test cases. It’s not a sensational replacement for a machine, because. The real encouragement here is that we can finally get the tools to improve the reach and productivity of existing healthcare systems. It’s an important tool for getting rid of inequity: making it accessible and affordable for everyone on the planet.”

With over 20 Babylon GP pools, 1,671 realistic written medical cases have been created. These cases included over 350 typical and atypical cases of disease symptoms. Each case was created by one doctor, validated by several other doctors, and confirmed to represent a real diagnostic case. At least 50 written cases (mean 159) were given and evaluated in a separate group of 44 Babylon GPs. Doctors listed the illnesses they considered most likely (on average, returning 2.58 potential illnesses with each diagnosis). They were measured for accuracy by the proportion of cases that included true disease in their diagnosis. Babylon’s AI underwent the same tests and used both the old correlation-based algorithm, which was created specifically for this study, and the new causal algorithm. In each test, the AI ​​was only able to report as many answers as the doctor’s.

The average doctor score was 71.40% (±3.01%), which ranged from 50 to 90%. The previous correlation algorithm performed as well as the average physician, achieving 72.52% (±2.97%). The score for the new causal algorithm was 77.26% (±2.79%), higher than 32 doctors, equal to 1 and less than 11.

“This AI will help me and other doctors reduce misdiagnosis, free up time, and focus on patients in need of care,” said Deputy Head of Medicine and GP Dr. Tejar Patel of Babylon. We are happy to help, and we look forward to when this type of tool becomes the norm and helps enhance our work.”

“Interestingly, we found that AI and doctors complement each other. AI scored better than doctors in more difficult cases, and vice versa,” said Dr. Saurabh Johri, Babylon’s chief scientist and author. Similarly, it is especially suitable for rare diseases that are more commonly misdiagnosed and often more serious.Switching from the use of correlation to about 30% accuracy in both rare and very rare conditions. Has improved.”

There is no need to change the underlying model of disease used by AI to improve accuracy. This is a benefit that applies to existing correlation algorithms, including those outside the healthcare environment.

Dr. Ciaran Lee, research author and former UCL honorary lecturer at Babylon, said: “Causal machine learning allows us to ask richer, more natural questions about medicine. This method has great potential to improve all other symptom checkers today. Healthcare and other It can be applied to many other problems as well, so causal AI is very impressive and universal.”

This technology paves the way for future partnerships between doctors and AI, speeding up doctors’ diagnoses, improving accuracy, freeing clinicians’ time, improving patient outcomes and patient experiences. .. It could augment the work of the clinician and keep driving better healthcare systems for the patient.

This new causal relationship algorithm Not yet in Babylon’s public app. It will only be released after further development and testing, and after meeting all necessary regulatory approvals in the UK and other markets where it is released.

Offered by Babylon