AI models improve the accuracy of diagnosing coronary artery disease

AI models improve the accuracy of diagnosing coronary artery disease

Several recent discoveries show that the accuracy of diagnosing coronary artery disease and predicting patient risk are being improved using artificial intelligence (AI) models developed by scientists in the Division of Artificial Intelligence in Medicine at Cedars-Sinai.

These advances, led by Piotr Slomka, PhD,Director of Innovation in Imaging at Cedars-Sinai and research scientist in the Division of Artificial Intelligence in Medicine and the Smidt Heart Institute, facilitate the detection and diagnosis of one of the most common and deadly heart diseases.

Coronary artery disease affects the arteries that supply blood to the heart muscle. If left untreated, it can lead to a heart attack or other complications such as cardiac arrhythmias or heart failure.

The condition, which affects approximately 16.3 million Americans ages 20 and older, is commonly diagnosed using single photon emission computed tomography (SPECT) and computed tomography (CT). However, the images produced during scanning are not always easy to read.

“We continue to show that AI can improve the quality of images and reveal more information, leading to more accurate disease diagnoses,” said Slomka, who is also a professor of medicine and cardiology and senior author of three recently published studies using AI to improve cardiac imaging.

Using AI to Improve Cardiac Imaging

The first study, published in the Journal of Nuclear Medicine, uses AI technology for cardiac imaging, helping to improve the diagnostic accuracy of SPECT imaging for coronary artery disease through advanced image corrections.

Attenuation correction is important in SPECT imaging, helping to reduce artifacts in cardiac images and making them easier to read and more accurate. However, it requires an additional CT scan and expensive hybrid SPECT/CT machines, which are essentially two scanners in one.

While CT attenuation correction has been shown to improve the diagnosis of coronary artery disease, it is currently only performed on a minority of scans due to additional scanning time, radiation, and limited availability of this expensive technology.

To overcome these obstacles, Slomka and his team developed a deep learning model called DeepAC to generate corrected SPECT images without the need for expensive hybrid scanners. These images are generated by AI techniques similar to those used to generate “deep fake” videos and can simulate high quality images obtained from hybrid SPECT/CT scanners.

The team compared the diagnostic accuracy of coronary artery disease with uncorrected SPECT images used in most places today, advanced hybrid SPECT/CT images, and new AI-corrected images in unseen data from centers that were never used in DeepAC training.

They found that AI created images that were nearly the same quality and enabled similar diagnostic accuracy as those obtained with more expensive scanners.

This AI model was able to generate DeepAC images in fractions of a second on standard computer software and could be easily implemented into clinical workflows as an automated pre-processing step.”

Piotr Slomka, PhD,Director of Imaging Innovation, Cedars-Sinai

Prediction of major adverse cardiac events

In the second study, published in the Journal of American College of Cardiology: Cardiovascular Imaging, the team showed that deep learning AI makes it possible to predict major adverse cardiac events such as death and heart attack directly from SPECT images.

Investigators trained the AI ​​model using a large multinational database that included five different locations with over 20,000 patient scans. It contained images depicting cardiac perfusion and movement for each patient.

The AI ​​model provides visual explanations for the doctors and highlights the images with the regions that contribute to a high risk of adverse events.

The team then tested the AI ​​model at two different locations with over 9,000 scans. They found that the deep learning model predicted patient risk more accurately than software programs currently used in the clinic.

“In the first study, we were able to show that important image corrections can be carried out with AI without expensive scanners,” says Slomka. "Second, we show that existing images can be put to better use - using images to predict the patient's risk of heart attack or death and highlighting the cardiac features that indicate this risk - to better inform clinicians about coronary heart disease."

“These results represent proof of principle for how AI can improve clinical diagnostics,” said Sumeet Chugh, MD, director of the Division of Artificial Intelligence in Medicine. “AI-powered improvements to SPECT imaging have the potential to improve the accuracy of diagnosing coronary artery disease while being performed significantly faster and cheaper than current standards.”

Reducing bias in AI models

The third study, published in the European Journal of Nuclear Medicine and Molecular Imaging, describes how an AI system can be trained to work well in all applicable populations - not just the population for which the system was trained.

Some AI systems are trained on high-risk patient populations, which can cause systems to overestimate the likelihood of disease. To ensure that the AI ​​model works accurately for all patients and reduces bias, Slomka and his team trained the AI ​​system with simulated variations of patients. This process, called data augmentation, helps better reflect the mix of patients expected to undergo the imaging tests.

They found that the models trained with a balanced mix of patients more accurately predicted the likelihood of coronary artery disease in women and low-risk patients, potentially leading to less invasive testing and more accurate diagnosis in women.

The models also resulted in fewer false positives, suggesting the system may be able to reduce the number of tests the patient undergoes to rule out the disease.

“The results suggest that improving training data is critical to ensuring that AI predictions better reflect the population to which they are applied in the future,” Slomka said.

Researchers are now evaluating these novel AI approaches at Cedars-Sinai and exploring how they could be integrated into clinical software and used in standard patient care.

The research was supported in part by the National Heart, Lung, and Blood Institute.

Source:

Cedars Sinai

Reference:

Shanbhag, AD, et al. (2022) Deep learning-based attenuation correction improves the diagnostic accuracy of cardiac SPECT. Journal of Nuclear Medicine. doi.org/10.2967/jnumed.122.264429.

.