Doctors often get early warning of impending heart failure, in part by detecting excess fluid in the lungs, and researchers at the Massachusetts Institute of Technology have developed a new machine learning tool that can help them. The algorithm can detect serious cases of the condition with high accuracy, and the researchers behind it hope it can also be transformed to help manage other conditions.
The study was conducted at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and is in conjunction with a number of other promising machine learning and artificial intelligence tools that are reshaping medical diagnostics. With the power of modern computing, these algorithms can look at medical imaging data and discover subtle but critical changes in the human condition that clinicians cannot see, opening up exciting possibilities.
This could mean a CT scan to detect missing cancer diagnoses, or years before doctors see signs of Alzheimer’s disease. The new study also uses artificial intelligence to analyze how electrostat image results can help doctors identify patients most at risk of heart failure by identifying left-chamber dysfunction, although the new study follows a similar path, albeit with different mechanisms.
Doctors use X-ray images of the lungs to assess fluid build-up in patients at risk of heart failure, the severity of the condition, known as pulmonary edema, and then determine the course of treatment. The trouble is that these assessments are often based on such subtle characteristics that they can lead to inconsistent diagnostic and treatment options.
To introduce machine learning, the team trained its algorithms on more than 300,000 X-ray images and reports written by their counterparts. This involves developing certain language rules to ensure that the data is consistently analyzed across a large sample.
“Our model can be interpreted by turning images and text into compact digital abstractions,” said Geeticka Chauhan, co-lead author of the paper. “We trained it to minimize the difference in presentation between X-ray images and radiology report texts and to use reports to improve image interpretation.”
Studies have shown that a new machine learning algorithm can classify cases of severe pulmonary edema with high precision. When testing it, the team asked the machine learning algorithm to analyze a single X-ray image and classify the severity of the edema, ranging from 0 (healthy) to 3 (very, very severe). The algorithm was able to diagnose the correct level of edema in more than half the time, but even more impressively, it was able to accurately diagnose level 3 cases in 90 percent of the time.
The researchers hope the tool will help doctors better manage heart problems, while edema is linked to a range of conditions, including sepsis and kidney failure, so the algorithm’s potential may be broader. Researchers are currently working to integrate the tool into the workflow of an emergency room at a Boston medical center in the coming months.