Artificial neural networks in cardiac care

One model that evolved from artificial intelligence is artificial neural networks (ANNs), often interchangeably referred to as neural networks. It is a mathematical or computational model that processes interconnected data (artificial neurons) to find a pattern in that data. In this process, you have input data, which goes through a connectionist approach to output data. The system adapts and learns through the multitude of data that flows through it. The result is expert decision making, or even a prediction system, with close to 100% accuracy. No wonder doctors have been using AI and expert systems to deliver better and more timely healthcare to their patients.

Artificial Intelligence in Medicine (AIM) has nicely summed up some of the areas they have applied this to. They are:

• Generation of alerts and reminders
• Diagnostic Support
• Criticize and Plan Therapy
• Information Retrieval Agents
• Recognition and interpretation of images.

In a study conducted in the late 1990s, researchers Lars Edenbrandt, MD, Ph.D., and Bo Heden, MD., Ph.D., of Lund University Hospital, Sweden, ventured to include 1,120 records ECG of patients with myocardial infarction. , and 10,452 records of normal patients. It was found that neural networks can use these input data and establish a relationship and pattern. This learning phase was internalized by the system and began to identify patients with abnormal ECGs with an accuracy 10% better than most physicians/cardiologists on staff.

But one point to keep in mind here is that technology like ANN cannot replace a doctor looking at various other factors and fine print before diagnosing a heart attack. It can simply be used as a means to supplement and speed up the diagnostic process.

Speaking of other factors in the determination of Heart Attacks, an interesting research paper had been published in a scientific journal of the Inderscience group, International Journal of Knowledge Engineering and Soft Data Paradigms (IJKESDP) under the name of “A computational algorithm for assessment of the risk of development of acute coronary syndromes, using analytical process methodology online” (Volume 1, Number 1, Pages 85-99, 2009). Four Greek researchers had ventured to develop a computational algorithm that evolved from a newer technique, namely Online Analytical Processing (OLAP). They used this methodology to build the foundation of a “Heart Attack Calculator.” The advantage of OLAP is that it provides a multidimensional view of the data, which allows you to discern patterns in a very large data set, which otherwise would have remained invincible. It takes numerous factors and dimensions into account when making an analysis. The research team obtained data from about 1,000 patients who have been hospitalized due to symptoms of Acute Coronary Syndrome. These data included details about their family history, physical activities, body mass index, blood pressure, cholesterol, and level of diabetes. This was then compared to another similar multidimensional data set from a group of healthy individuals. All of these data were used as inputs to the OLAP process, to explore the role of these factors in cardiovascular disease risk assessment. At various levels of the factors, intelligence could be collected to be used as a combination of dimensions, for future diagnosis of the extent of risk.

The ANN is more of a “teachable software”, which absorbs and learns from data input. When calculated correctly, even at a rapid pace with a tried and tested algorithm, it develops patterns within the input data, or a combination of multiple data dimensions or factors, against which a given situation can be compared and state a forecast. .

In 2009, Mayo Clinic researchers studied 189 patients with device-related endocarditis diagnosed between 1991 and 2003. Endocarditis is an infection that affects the valves and sometimes the chambers of the heart, often caused by implanted devices. in the heart. Mortality due to infection could reach 60%. The diagnosis of this infection required a transesophageal echocardiography, which is an invasive procedure that involves the use of an endoscope and the insertion of a probe down the esophagus. Needless to say, this was a risky, uncomfortable, and expensive procedure. The Mayo researchers entered data from these 189 patients into the ANN and put it through three separate “trainings” to learn how to assess these symptoms. By being tested with different sample populations (only known cases and then a general sample of a mix of known and unknown cases), the best-trained ANN was able to identify endocarditis cases very effectively, thus eliminating the need for such a procedure. invasive. .

As modern eHealth is becoming more and more data-centric, access to relevant patient data is becoming more and more convenient. AI and expert systems with their ANN and computational algorithms have tremendous opportunities to accelerate diagnosis and perform patient care with speed and increasing precision. As AI advances, it will be interesting to see how it makes its mark in cardiovascular, neurological, pulmonary and oncological diagnosis and care.

I invite you to share your experiences and thoughts on how AI has affected various disciplines within medicine, most importantly within acute care.