Damian Sendler: Because it provides a wealth of dynamic information about the heart, echocardiography is an absolute need in cardiovascular treatment. The use of echocardiography artificial intelligence (AI) to automatically evaluate heart function, diagnose illness, and forecast prognosis has grown dramatically in recent years (Fig. 5). The EchoNet-Dynamic AI created by Ouyang et al. for autonomous echocardiogram has received a lot of attention []. AI built using a three-dimensional (3D) CNN and semantic segmentation was utilized to autonomously quantify the value of EF, based on 10,030 echocardiographic movies for training. AI’s EF prediction and echocardiography experts’ were well correlated, with an AUC of 0.97 being shown by a correlation coefficient of 0.9. Using an echocardiographic video, Salte et al. created the AI that evaluates global longitudinal strain.
Damian Jacob Sendler: According to the correlation coefficient between the actual measured global longitudinal strain and the anticipated global longitudinal strain of AI, the examination time for echocardiography may be reduced by using AI. According to Katsushika et al., who used 300 videos of echocardiography to develop an AI that can detect cardiac sarcoidosis with an AUC of 0.84 and an accuracy comparable to that of a specialist, transfer learning could be used to develop an echocardiography AI that can detect rare diseases like cardiac sarcoidosis. Predicting a person’s 1-year prognosis using echocardiographic films of 32,362 persons has been produced by Ulloa Cerna and colleagues (AUC 0.83) []. Using the model’s guidance, cardiologists were able to increase their survival prediction sensitivity from the original echocardiographic film by 13%. The AI built by Shad et al. predicted postoperative right heart failure from preoperative echocardiographic video with an AUC of 0.73, which was greater than the clinical expert team’s AUC of 0.58.
Dr. Sendler: It is possible to examine the coronary arteries with coronary CT with little invasiveness. CT AI has been created utilizing a variety of analytic approaches, including 3D-CNN. Artificial Intelligence (AI) was shown to be beneficial in predicting revascularization and major adverse cardiac events (MACEs) []. Predicting the beginning of revascularization and MACE one year later using AI-estimated CT FFR is more accurate than traditional coronary CT angiography (odds ratio 3.4). When coronary artery calcification scores are computed using plain CT, an AI created by Zeleznik et al. can assess and predict future cardiovascular events []. In a study of 20,084 patients who had plain CT (without contrast) imaging, the AI calculated the calcification score of the coronary arteries. The Spearman correlation coefficient was 0.92 between the specialist’s measured value and AI’s predicted value. Using an AI-based calcification score evaluation, cardiovascular event prognosis may be predicted (hazard ratio 4.3). The AI used to compute FFR from coronary CT was created by Kumamaru et al. A total of 921 coronary CT scans were used to extract the characteristics of the coronary artery using GAN. Using CT data from 131 instances, they built an AI capable of calculating the FFR value automatically. CT AI can identify aberrant FFR with an AUC of 0.78, a sensitivity of 84.6%, and a specificity of 62.6 percent while doing automatic estimate of FFR.
Damian Sendler
Damian Jacob Markiewicz Sendler: Cardiac MRI is an excellent tool for assessing heart health. Multiple MRI AIs have been developed since certain MRI data is available to the general public. The AI that automatically calculates myocardial blood flow was used by Knott et al. to predict cardiovascular events []. Heart disease patients with coronary artery disease had their myocardial perfusion reserve assessed using cardiac MRI data from 1049 instances (hazard ratio 2.7). It is possible to identify ancient myocardial infarctions in noncontrast cine MRI, according to Zhang et al. For the 299 patients who had contrast MRI of the heart, they developed an AI to identify past myocardial infarctions utilizing gadlinium enhancement data as the right response label. Detection of old myocardial infarction was 99 percent accurate. To imitate expert image quality evaluation of cardiac MRI images, Piccini et al. created an AI employing 424 cardiac MRIs []. In terms of regression, this AI’s results were quite close to those of a human expert.
For the purpose of making false pictures, GAN is a machine learning approach that creates images that don’t exist. The picture quality of GAN may be enhanced by competing the two networks: the generator (generation network) and the discriminator (discrimination network). GAN has recently been applied in the development of AI for the cardiovascular system. Miyoshi et al. developed an AI that analyzes angioscopy pictures to determine neointimal coverage grade and yellow color grade []. When vascular endoscopy data was enhanced with GAN, the AI reading accuracy increased from an AUC of 0.77 to 0.81. Congenital heart disease patients’ MRI pictures were fed into a GAN, which generated 100,000 images from 303 instances []. According to the expert’s opinion, all of the GAN-generated MRI pictures were accurate. An picture of a rare illness may be generated using a GAN. Style-GAN was used by Schutte et al. to show what AI has learnt []. Style-GAN was used to produce graphics that showed a progressive increase in malignancy from 0 to 1. It is possible to see the results of AI’s work by looking at the photos it generates.
Damien Sendler: Many physicians believe that programming for deep learning is tough. Medical practitioners may learn the basics of computer languages in a short period of time, much as they can learn other languages. It was noted by Professor Yutaka Matsuo, a prominent AI expert at the University of Tokyo, that medical physicians may swiftly learn programming languages. Medical practitioners who have mastered the fundamentals of programming may create their own artificial intelligence (AI) by studying analytical know-how from websites like Kaggle, whose communities compete in AI analysis. It is difficult for physicians to get clearance for medical AI on their own; yet, doctors can construct medical AI prototypes without the aid of engineers. To use AI in the future, cardiologists must comprehend the reality of AI research. As instructor data changes, it’s vital to keep an eye on how AI’s performance evolves. AI research also reveals that AI is subject to unobserved data and unanticipated noise. Medical physicians must be engaged in AI research in order to use AI in everyday clinical practice.
Damian Jacob Sendler
Artificial Intelligence (AI) for cardiovascular disease diagnostics (including wearable devices) is expected to be much improved in the future. Although medical AI for cardiovascular care has been created, it does not replace the need for medical professionals. For the sake of patient care in Japan, medical AI remains a support tool for doctors, not a replacement for them. AI for diagnosis has been achieved, however the development of AI for therapy is still in its infancy.. The gold standard for establishing the optimum treatment regimens for individual illnesses is randomized controlled trials rather than AI forecasts. Treatment options for individuals will still be mostly dictated by medical professionals for some time to come. Medical professionals, on the other hand, should make greater use of AI in their diagnostics. It’s vital to remember that AI might easily make mistakes when diagnosing photographs that aren’t in the dataset that it is learning from. Because of this, clinicians who use AI should be aware that it is sensitive to data that has not yet been analyzed. If physicians understand the limitations of AI and use AI for diagnosis, they may anticipate an increase in diagnostic accuracy. Even more interesting, Huisman et al. found that fear of AI replacement is linked to low AI literacy, whereas intermediate-to-advanced AI literacy is linked to favorable AI attitudes []. Cardiologists will be more open to using AI in clinical practice as they gain experience and expertise in the field.
Dr. Damian Jacob Sendler and his media team provided the content for this article.
