Pneumonia is widely considered to be a fatal disease in the U.S. Not only does the illness put millions of Americans in hospitals annually but also claims at least 50,000 casualties. In essence, doctors primarily conduct a chest X-ray once a patient is suspected to suffer from the disease. With the recent technological advancement in almost every conceivable industry, Stanford researchers have made significant progress by developing an algorithm that can allegedly diagnose pneumonia on X-rays much quicker than seasoned radiologists.

 

The researchers believe that the algorithm has the unique advantage of learning from thousands of chest X-rays and provide their corresponding diagnoses. According to Pranav Rajpurkar, a key member of the Stanford Machine Learning Group, radiologists hardly get the chance to learn valuable knowledge from thousands of past diagnoses and subsequently come up with reliable patterns that can lead to successful diagnoses.

 

How It Works

 

Known as CheXNet, the algorithm is adept at diagnosing over 13 medical-related conditions such as pneumothorax and emphysema. Through the crafted expertise of the team, the algorithm operates based on a precise public dataset from the National Institute of Health, which is sufficiently backed with over 100,000 chest X-ray pictures titled with corresponding conditions. As a testament to their proficiency, the dataset has been released in conjunction with the initial diagnosis algorithm to wade off any insecurity that might crop up.

 

To test the efficiency of the algorithm, Rajpurkar and his team made the bold decision to mark a sample of 420 images proven to possess strains of pneumonia. Through this sample, the team brilliantly came up with the algorithm that could initially detect ten conditions. A week later, the algorithm had extended its scope, and could successfully diagnose over 14 infections.

 

After its successful trial phase, CheXNet has unanimously won over the hearts of radiologists from locally and beyond based on its unrivaled performance. It is definitely poised to make a huge difference in the diagnoses of illnesses like pneumonia.

 

Source: https://www.smithsonianmag.com/innovation/algorithm-diagnose-pneumonia-180967327/