Researchers at the Stanford University’s Machine Learning group have developed an Algorithm which presents diagnosis for Pneumonia. The Algorithm is based on a Deep Learning Algorithm called Convolutional Neural Networks. The task of this AI algorithm was to analyse images of Chest X-Rays of patients and diagnose one of the 14 types of pneumonia conditions observed in patients. The Algorithm is reportedly better at identifying Pneumonia better than a Radiologist working alone. The algorithm demonstrates an 85% accuracy.

The paper, CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning, published by the researchers is available publicly at arXiv.org, a scientific pre-publication website.

Pneumonia is a particularly complicated disease to diagnose. The researchers came across the public dataset released by the National Institutes of Health Clinical Center which consists of 112,120 frontal-view chest X-ray images labeled with up to 14 possible pathologies. The Algorithm uses a 121-Layer Convolutional Neural Network from the Deep Learning branch of Artificial Intelligence. The experiment involved four Radiologists to manually annotate the X-Ray images from the dataset to evaluate the performance of the Algorithm. Researchers found that the algorithm performs better than an average radiologist at both sensitivity and specificity when diagnosing the said conditions.

Complementary to the algorithm, the researchers have also developed a tool which generates an image similar to a heat map. Instead of the heat spectrum, the map highlights the areas of the X-Ray which are more likely to represent a pneumonia condition. The tool intends to reduce the number of missed cases encountered by doctors.

Pranav Rajpurkar, a graduate student in the Machine Learning Group at Stanford and co-lead author of the paper while discussing the problem said, “Interpreting X-ray images to diagnose pathologies like pneumonia is very challenging, and we know that there’s a lot of variability in the diagnoses radiologists arrive at. We became interested in developing machine learning algorithms that could learn from hundreds of thousands of chest X-ray diagnoses and make accurate diagnoses.”

The researchers collaborated with Matthew Lungren, MD, MPH, assistant professor of radiology at the School of Medicine at Stanford who employed four radiologists to independently annotate 420 images of the available dataset which would be compared to the performance of the Algorithm. The Algorithm evaluates against the sensitivity (which measures the proportion of positives that are correctly identified as such) and specificity (which measures the proportion of negatives that are correctly identified as such) to calculate the accuracy.

“The motivation behind this work is to have a deep-learning model to aid in the interpretation task that could overcome the intrinsic limitations of human perception and bias, and reduce errors. More broadly, we believe that a deep-learning model for this purpose could improve healthcare delivery across a wide range of settings.” explained Lungren.

The Machine Learning Group at Stanford is led by Andrew Ng who is a well-known figure in the Machine Learning and AI community. The group hopes to tackle much broader and complicated problems in the future to improve healthcare and other fields. In areas with a lack of medical expertise, such algorithms could augment doctors to delegate their diagnoses to machines so that they can utilize their time in the treatment of their patients.