What is the risk of sudden death within 10 years? The first neural network algorithm tells you

Sudden cardiac death (SCD) remains the leading cause of death worldwide, with an incidence of 50-100 per 100,000 people in the general population in Europe and North America, accounting for 15-20% of all deaths. Patients with coronary artery disease are at the highest risk of sudden cardiac death from arrhythmia (SCDA).

Therefore, there is an urgent need to develop personalized, accurate, and cost-effective arrhythmia risk assessment tools to alleviate this huge public health and economic burden.

Recently, a team led by researchers at Johns Hopkins University has developed a new artificial intelligence-based method that can predict more accurately than doctors whether and when a patient may die from cardiac arrest. The technology, based on the patient's cardiac imaging data and other background, will revolutionize clinical decision-making and improve survival rates for sudden and fatal arrhythmias.

The related research was published in the recent issue of Nature Cardiovascular Research.

“Sudden cardiac death, caused by arrhythmias, accounts for 20 percent of all deaths worldwide, but we know very little about why it happens or how to tell who is at risk,” said Natalia Trayanova, professor of biomedical engineering and corresponding author of the paper. “Some patients may be at low risk for sudden cardiac death and may not need an automated external defibrillator (AED), while some high-risk patients may die in the prime of their lives if they don’t get the treatment they need in a timely manner. What our algorithm can do is determine who is at risk for cardiac death and when it will happen, allowing doctors to decide exactly what needs to be done.”

To our knowledge, this is the first team to use a neural network to create a personalized survival assessment for each heart attack patient. These risk measures provide a high degree of accuracy for sudden cardiac death within 10 years, as well as the time when it is most likely to occur.

The researchers named this deep learning-based tool the Survival Study of Cardiac Arrhythmia Risk (SSCAR).

Figure | SSCAR schematic (Source: Nature)

In current clinical cardiac image analysis, doctors only extract simple scar features, such as volume and mass, and fail to fully utilize key data in the relevant images.

"These images carry critical information that doctors can't access," says first author Dan Popescu, a former Johns Hopkins doctoral student. "The scar can be distributed in different ways, and it tells a lot about a patient's chance of survival. It's just that the information is hidden."

To do this, the research team first used contrast-enhanced cardiac magnetic resonance images to visualize the distribution of scars in 156 real patients with cardiac magnetocardiomyopathy at Johns Hopkins Hospital to train an algorithm to detect patterns and relationships that are invisible to the naked eye.

Figure | SSCAR detected high risk in the red circled heart (Source: Johns Hopkins University)

The team also trained a second neural network using ten years of standard clinical patient data, which included 22 factors such as age, weight, race, and prescription drug use.

These parameters are then learned directly from CMR images and clinical factors using a deep neural network to optimally model the survival data, produce highly personalized survival probability predictions, and derive patient-specific survival curves.

The researchers then validated the algorithm in tests on an independent cohort of patients from 60 medical centers in the United States with different histories of heart disease and different imaging data. The results showed that the algorithm's predictions were much more accurate than those of doctors, and the results showed that the system could be widely used everywhere.

Notably, the overall design of the custom neural network used in SSCAR took multiple steps to ensure the relevance and interpretability of the resulting features. Interpretability of AI algorithms is critical to their widespread adoption, and concerns surrounding it are particularly prevalent in the healthcare sector.

“This has the potential to significantly impact clinical decisions about arrhythmia risk and represents an important step in bringing patient trajectory prediction into the AI ​​era,” said Trayanova, co-director of the Alliance for Innovation in Cardiovascular Diagnosis and Treatment. “It exemplifies the trend toward the convergence of AI, engineering, and medicine as the future of healthcare.”

The team is currently working on building algorithms to detect other types of heart disease. According to Trayanova, deep learning concepts can also be developed for other medical fields that rely on visual diagnosis.

References:

https://www-nature-com-443.webvpn.bjmu.edu.cn/articles/s44161-022-00041-9

https://hub.jhu.edu/2022/04/07/trayanova-artificial-intelligence-cardiac-arrhythmia/

Source: Academic Headlines