Damian Sendler: The necessity for physicians to have a working knowledge of artificial intelligence (AI) is growing in the therapeutic context. An examination of the fundamentals of artificial intelligence and the status of cardiovascular artificial intelligence is the topic of this study. There are a variety of AI systems for analyzing cardiovascular examinations, including X-rays and ECGs, echocardiograms, CT scans, and MRIs. The diagnostic and prognostic prediction accuracy of cardiovascular AI is quite good. Furthermore, it is capable of detecting anomalies that cardiologists have previously had difficulty detecting.
Damian Jacob Sendler: Randomized controlled experiments began to be published to validate the effectiveness of cardiovascular AI. Cardiovascular artificial intelligence (AI) is on the verge of becoming commonplace in clinical practice. In cardiovascular treatment, many forms of medical AI will be deployed; nonetheless, physicians will not be replaced by them. To better serve their patients, cardiologists must have a thorough understanding of the benefits and shortcomings of medical AI.
Dr. Sendler: A number of research on medical AI have been published in the last several years [,]. Image interpretation in the domains of radiology and pathology, in particular, is becoming more dependent on artificial intelligence (AI). Neuroprosthesis for stroke patients employing a brain–computer interface has been reported as a novel AI study in addition to diagnostic imaging [,]. AI experiments in the cardiovascular sector have also been described []. AI is being developed for a wide range of medical tests, including X-ray, ECG, echocardiography, CT, and MRI, among others (Fig. 1). Using machine learning approaches, Karwath et al. built an AI that can identify heart failure patients who react to beta-blockers. Using an ECG during normal sinus rhythm, Attia et al. developed an AI that accurately predicts future paroxysmal atrial fibrillation based on an ECG AI that detects decreased left ventricular contractility.
Damian Sendler
ECG AI was studied by Yao et al. in a randomized controlled experiment and shown to be useful in clinical practice. Medical recommendations will advocate the use of ECG AI if sufficient evidence supporting its efficacy can be found. Wearable gadgets, such as the Apple Watch, have previously been shown to be useful in identifying arrhythmias; in Japan, for example, their use is increasing [].
As a result of deep learning, AI can now recognize images and be considered to have developed “eyes.” Artificial Intelligence (AI) is now in the process of developing hands, ears, and lips, among other features and capabilities. In an effort to produce robot arms that can do delicate tasks like human hands, there has been a worldwide competition. Voice recognition and natural language processing are being worked on by engineers to make it easier for people to converse. The “ears” and “mouths” of smart speakers are on the verge of becoming fully human-like. AI’s “hands,” “ears,” and “mouths” — in addition to its “eyes” — might be useful in a variety of medical settings. The use of artificial intelligence (AI) in clinical settings is expanding, and the necessity for doctors to comprehend AI is increasing as well.. This review discusses the fundamentals of artificial intelligence and the present and future state of cardiovascular artificial intelligence.
Damian Jacob Markiewicz Sendler: Recent years have seen an increased focus on deep learning in machine learning. [Deep learning is based on neural networks that simulate human brain neural circuits]. An input layer receives data, a hidden layer (or intermediate layer) processes the weights that flow from the input layer, and an output layer outputs a result from the neural network (Fig. 3). The number of hidden layers in neural networks is multiplied in the “deep learning” model in order to increase accuracy. Increased complexity may be handled by increasing the number of hidden layers. Unlike standard machine learning, deep learning can automatically derive data characteristics from a given set of data without of having to specify them. High-performance recognition was previously impossible with manually run machine learning. Deep learning for image identification has come a long way in the last few years. The standard for image recognition is a convolutional neural network (CNN).
Using CNNs, we can extract and create networks that react to critical visual inputs in the same way as the visual cortex of animals responds. The study of dynamic and static pictures at great depth is becoming more popular. The medical profession is also benefiting from emerging deep learning approaches, such as generative adversarial network (GAN) and transfer learning [,]. Artificial Intelligence (AI) systems that can understand natural language have been created, and the field of language AI is attracting researchers’ interest [,]. There are many unanswered difficulties, including as ethical and legal concerns around the usage of AI. Advances in AI technology will soon be used to the medical area as well.
Damian Jacob Sendler
It is common in cardiology to do X-rays of the chest as part of a diagnostic evaluation. The application of X-ray AI for the detection of pneumonia was described by Kermany et al. []. Transfer learning may be used in the medical industry to diagnose pneumonia, as shown by chest X-rays. Learning a model in one area and transferring it to another is known as “transfer learning” (Fig. 4). For their initial AI model, they used 12 million photos that had been manually labeled []. Using X-rays from 5232 patients, the AI model was trained to diagnose pneumonia with an AUC of 0.97 and a sensitivity and specificity of 93.2 percent and 90.1 percent, respectively. The prognosis could only be predicted by AI based on chest X-rays, according to a study by Lu et al. Based on the results of chest X-rays taken from 52,320 patients, a prognostic model was developed that divides the mortality risk into five categories: very low, low, moderate, high, and very high. A prognosis based only on chest X-rays and AI was found to be 1, 1, 14, 1, 7, 2, 6, and 4, with background-adjusted risk mortality ratios. According to Toba et al., 657 individuals with congenital heart disease were used to construct the AI that uses chest X-ray data to infer hemodynamics [].
Damien Sendler: The pulmonary-to-systemic flow ratio measured with a catheter and the AI-derived value determined from X-ray data had a correlation coefficient of 0.68. AUC of 0.88 for X-ray data-assisted AI was shown to be very accurate in the identification of a high pulmonary-to-systemic flow ratio, compared to an AUC of 0.78 for the experts. Using chest X-ray scans, an AI developed by Matsumoto et al. was able to distinguish between heart failure and normal. With the use of ImageNet’s VGG16 and transfer learning, an AI was able to accurately identify heart failure in 638 chest X-rays with an accuracy rate of 82%. In terms of sensitivity and specificity, the results were 75% and 94%.
The electrocardiogram (ECG) is a noninvasive and straightforward diagnostic often used in cardiology. It has been widely utilized in clinical practice for a long time, and automated diagnostics of arrhythmia and ST abnormalities are already possible. ECG AI may be able to detect anomalies that were previously thought impossible to detect. “Atrial fibrillation in sinus rhythm may be predicted by electrocardiographic artificial intelligence,” as described by Attia et al. For the prediction of atrial fibrillation, the AUC was 0.87, which shocked the physicians; the specificity was 79.5 percent and the accuracy was 79.5 percent. Another study found that ECG AI can identify changes in ejection fraction (EF) from the ECGs of 44,959 individuals and demonstrate an AUC of 0.93, sensitivity of 86.3 %, and specificity of 85.7 percent []. In a randomized controlled study, Yao et al. investigated the ECG AI’s ability to identify EF decline []. In order to compare the diagnosis rate for identifying EF decrease, they randomly allocated 22,641 instances to one of two groups (one with and one without ECG AI). The detection rate of EF decline rose by around 30 percent in the group that used ECG AI. The ECG AI developed by Goto et al. may be used to diagnose cardiac amyloidosis []. []. The AUC was 0.91 in the model based on the ECG data of 3191 patients. When this approach was used in conjunction with echocardiography, the AUC increased to 0.96. ECG data from 29,859 patients was used to construct an AI that correctly identified aortic regurgitation with an AUC of 0.80 []. The advancement of ECG AI in recent years has been fast.
Dr. Damian Jacob Sendler and his media team provided the content for this article.
