Pneumonia is an inflammatory condition that affects around 450 million people across the world each year, causing around 4 million deaths annually. Now, scientists at Stanford University have developed artificial intelligence (AI)-based software that can diagnose pneumonia better than doctors ever could.
The new AI software is called CheXnet and is a kind of neural network that works by processing images via a set number of parameters. Led by Stanford adjunct professor Andrew Ng, the research team first had to train the neural network using over 110,000 chest X-ray images that covered 14 different diseases, one of which was pneumonia. After just one month of training the AI was able to detect this condition faster than any previous computer-based method used.
When the AI software went up against four human, Stanford radiologists they proved more accurate than them too in diagnosing pneumonia. Not only does the AI produce more accurate results it also makes the whole process of diagnosing the condition much simpler. Using this method, doctor’s input a patient’s lung X-ray into the computer and automatically get given the probability of pneumonia being present in the lung. A color map is also produced that demonstrates the level of the infection in the area. Using this data doctors can develop the most effective treatment plan for the patient.
The team is hoping this new method will be quickly adopted and used in various medical settings all around the world. At the moment, as many as two thirds of the global population have no access to accurate radiology diagnostic tools. With a little push, this tool could improve the delivery of healthcare for millions of people where skills and tools are scarce.
IEEE Spectrum’s Tekla S. Perry knows only too well how important this AI is. Her own son went to the ER at Stanford Medical Center twice with an “extremely high fever and cough”. Both times the X-rays showed no sign of pneumonia. Only when a routine reevaluation was carried out of X-ray images did doctors suspect he could (and did) have pneumonia.
According to a research paper published in October 2010, as many as “72% of patients were misdiagnosed with pneumonia upon readmission to the same hospital.” So, it looks like that’s another point for AI then.