Who Is Better At Medical Diagnostics: Doctors Or AI?

Who Is Better At Medical Diagnostics: Doctors Or AI?

Even though artificial intelligence has not yet reached the scales described in science fiction, you can still be pleasantly surprised with its achievements to date. For instance, in the recent years there have been reports saying that AI began to diagnose some diseases better than doctors. Is it really so and what does it mean for future medicine?

It is already proved that AI outperforms humans in diagnosing some types of cancer, Alzheimer's and Parkinson diseases and also at diagnosing pneumonia and some heart diseases. This happens because AI is able to proceed large amounts of information and unbiasedly compare it which cannot be done by humans. There are more and more competitions taking place between AI and doctors trying to make a more accurate diagnosis. So, it becomes clear that AI has great potential in this field which is growing with time.

There are many projects that focus on diagnostics of different diseases using AI. However, not all of them cope with their task better than humans. Below you can see the brightest researches which showed benefits of AI use and how it outperforms doctors.

Where AI has already outperform doctors?

1. Skin cancer diagnostics


In May 2018 there was an experiment in which 58 dermatologists compete with convolutional neural network. It turned out that doctors correctly diagnosed cancer in 86.6% of cases and AI – in 95%. For training the system developers used more than 100.000 pictures of this disease. According to the team, the system is able to diagnose skin cancer at early stages before it starts spreading.

2. Cancer and heart failure diagnostics


Private research university in Cleveland (Ohio, USA) has its lab Madabhushi where researchers developed a machine vision system for diagnosing different types of cancer (breast, prostate, head and neck), epilepsy and heart failure.
Their system correctly predicted heart failure in 97% of cases while doctors could only diagnose 74% of cases. There was also an experiment about diagnosing cancer nodules and while human radiologists can flag up to half of all nodules as "suspicious" or "indeterminate", about 98% eventually turn out to be benign. Computer was 8% better than humans in this experiment.

3. Pneumonia diagnostics


Stanford researchers developed a deep machine learning algorithm which assesses chest X-ray images and can detect pneumonia. The system can diagnose up to 14 types of medical conditions. Competing with 4 experienced radiologists their system won a confident victory.

4. Early diagnostics of cardiovascular diseases


Oxford researchers do not lag behind their American colleagues. Now they can be proud of having developed a system that can analyze heart scans and predict some diseases and possibility of heart attack more accurately than doctors who cannot always diagnose it in time and sometimes advise patients to take an unnecessary operation. The heart disease technology should be available to NHS hospitals in England for free this summer.

Despite such positive results, AI cannot replace doctors in the near future and there are many reasons for this, one of which is ethical issues. If AI will be the only one to diagnose then there is an issue of defining who is responsible for its decision. However, doctors admit that AI has powerful benefits and they see it as a decision-making consultant which can quickly and impartially process huge data amounts and draw a conclusion. However, the last word will be made by humans. In general, there is no sharp need to give responsibility for making decisions to a system. Even as a consultant, AI greatly influences the healthcare system and improves the accuracy and speed of diagnosis reducing the number of mistakes and costs of medical institutions.