Ways ChatGPT (and other generative AI apps) Will Change Healthcare Forever, For the Better
April 1, 2023Performance metrics of machine learning methods
April 1, 20231994 – 1998
In the mid-90s, artificial intelligence was decisive in making therapeutic and diagnostic decisions, establishing better results compared to expert professionals in the cardiovascular and ophthalmological areas...
It was well established in the mid-’90s, the processing and interpreting of images to analyze skin lesions (Cheung, 1994) or retinal photography (Gardner et al., 1996).(1) In 1994, a clinical study by D. Cheung was published to digitalize melanocytic lesions and to use the symmetric distance to improve the measurement of asymmetries of the skin lesions, which gave more accuracy in the diagnosis of malignant lesions.(2) A study performed in 1996 by G. Gardner compared neural networks against the conventional digital image analysis by ophthalmologists screenings. The study demonstrated that artificial analysis achieved good accuracy in detecting diabetic retinopathy.(3)
Artificial intelligence can be physical, such as robotic surgery, or virtual, relating to digital image manipulation, neural networks, and machine and deep learning. (1) Artificial neural networks have also been used to handle a variety of obstacles in different areas of cardiovascular medicine. Coronary artery disease, electrocardiography, cardiac image analysis, and cardiovascular medicine dosage are fields of interest.
Using an artificial neural network, a research group evaluated the relevance of coronary artery disease. Based on 23 noninvasive factors, they trained a personal computer-based artificial neural network to identify patients with severe coronary artery disease, with a positive predictive accuracy of 0% and a negative predictive accuracy of 92%. (4)
An area of research that takes more significant advantage of artificial intelligence corresponds to the use of artificial neural networks for diagnosing myocardial infarction through ECG. Cardiovascular studies for anterior and inferior infarctions showed that the artificial neural network’s sensitivity was 81% and 78%. However, the conventional program’s sensitivity showed 68% and 66%, respectively. (6)
Regarding the analysis of cardiac images, an important study carried out by Cios et al. used echocardiography images through neural networks and established areas of diagnostic interest. The investigators found specificities of >70% for all diagnostic classes, such as hypertrophic cardiomyopathy and posterior myocardial infarction. (7)
An area of research that takes more significant advantage of artificial intelligence corresponds to the use of artificial neural networks for diagnosing myocardial infarction through ECG. Cardiovascular studies for anterior and inferior infarctions showed that the artificial neural network’s sensitivity was 81% and 78%. However, the conventional program’s sensitivity showed 68% and 66%, respectively. (6)
Regarding the analysis of cardiac images, an important study carried out by Cios et al. used echocardiography images through neural networks and established areas of diagnostic interest. The investigators found specificities of >70% for all diagnostic classes, such as hypertrophic cardiomyopathy and posterior myocardial infarction. (7)
Finally, establishing effective treatments is an important goal of clinical decision-making. “Hypernet,” an artificial neural network-based system developed by Poli et al., uses inputs from a 24-hour sphygmomanometer to diagnose and analyze therapeutic interventions in outpatients. A test group of 35 patients compared the treatments recommended by Hypernet with the recommendations of experienced professionals. In this comparison, Hypernet achieved a sensitivity of 92% and a specificity of 96% when evaluating diagnostic and therapeutic effects. (5)
Finally, establishing effective treatments is an important goal of clinical decision-making. “Hypernet,” an artificial neural network-based system developed by Poli et al., uses inputs from a 24-hour sphygmomanometer to diagnose and analyze therapeutic interventions in outpatients. A test group of 35 patients compared the treatments recommended by Hypernet with the recommendations of experienced professionals. In this comparison, Hypernet achieved a sensitivity of 92% and a specificity of 96% when evaluating diagnostic and therapeutic effects. (5)