Detecting Heart Attacks with Computed Tomography
Applying machine learning algorithms and texture analysis to CT images can identify heart attacks.
A combination of texture analysis and machine learning can detect heart attacks, also known as myocardial infarctions, on cardiac computed tomography (CT) images -- even those not visible to the eyes of trained radiologists.
In a recent proof-of-concept study in the journal Investigative Radiology, researchers applied machine learning algorithms and texture analysis (which objectively quantifies the texture of radiological images not visible to the naked eye) to a specific type of CT scan called non-contrast-enhanced low-dose cardiac CT.
The researchers compared patients who had previously been diagnosed with either acute or chronic myocardial infarction to controls with no cardiac abnormalities. Using non-contrast-enhanced low-dose cardiac CT images, they performed texture analysis and applied machine learning classifiers including an artificial neural network employing deep learning. Additionally, two experienced radiologists previously uninvolved in the subjects' care visually assessed the images for signs of myocardial infarction.
The results indicated that certain texture analysis features, combined with machine learning algorithms, can identify patients with myocardial infarction on non-contrast-enhanced low-dose cardiac CT images with high accuracy and sensitivity.
What’s more, the experienced radiologists were not able to identify myocardial infarction in these images by visual assessment alone. So, texture analysis and machine learning show promise for unveiling information not visible to the naked eye on non-contrast-enhanced low-dose cardiac CT images.