According to a study in Radiology published by Science Daily, Patients admitted to the hospital with severe chest discomfort may benefit from using artificial intelligence.
Márton Kolossváry, M.D., Ph.D., a radiology research fellow at Massachusetts General Hospital in Boston, built the deep learning AI model, the first of its kind to use chest X-rays to predict whether patients experiencing acute chest discomfort require immediate medical assistance.
Acute chest pain syndrome can produce a wide range of symptoms, including chest tightness, burning, discomfort, and acute pain that extends to the back, neck, shoulders, arms, or jaw. It’s conceivable that you’re having breathing difficulties.
Acute chest pain syndrome affects around 7 million people in the United States annually. Less than 8% of people with acute chest pain have one of the three most common cardiovascular causes: acute coronary syndrome, pulmonary embolism, or aortic dissection.
Given the potentially catastrophic nature of cardiovascular and pulmonary disorders and the limited specificity of clinical tests such as electrocardiograms and blood tests, diagnostic imaging is widely used, even though it frequently produces false negative results.
It is critical to swiftly triage persons with a shallow risk of these catastrophic illnesses to relieve the load on emergency departments generated by a large volume of patients and a scarcity of hospital beds.
Deep learning AI can evaluate X-ray pictures for illness indicators when adequately taught. Dr. Kolossváry and colleagues built a deep learning model to predict whether a patient with acute chest pain syndrome would have a coronary event, a pulmonary embolism, an aortic dissection, or die within 30 days.
The study included patients with acute chest pain syndrome treated at either MGH or Brigham and Women’s Hospital in Boston between January 2005 and December 2015. The study included 3,329 males and 5,750 women (median age 59). The deep-learning model was trained on 23,005 chest X-rays to predict the 30-day composite endpoint of the acute coronary syndrome, pulmonary embolism, aortic dissection, and all-cause death.
When paired with demographic parameters like age and gender and traditional clinical indicators like d-dimer blood tests, the deep-learning technique considerably enhanced the accuracy with which these adverse outcomes could be anticipated. The model’s diagnostic effectiveness was maintained independent of the subject’s age, gender, race, or ethnicity.
At a sensitivity threshold of 99%, our model was able to defer further testing in 14% of patients. Still, a model containing patient age, gender, and biomarker data was only able to do so in 2% of cases.
“We were able to develop more accurate predictions regarding patient outcomes than a model that utilizes age, sex, troponin, or d-dimer information by utilizing an automated deep learning model to analyze the first chest X-ray photographs of these patients,” Dr. Kolossváry said.
Dr. Kolossváry speculates that in the not-too-distant future, an automated model may scan background chest X-rays to determine which individuals require immediate medical attention and which may be safely discharged from the emergency room.
