AI beat cardiologists
Neural network analyzes MRI of the heart more effectively than experts
Polina Gershberg, Naked Science
An international group of scientists from research centers in the UK and China has found out how good artificial intelligence is at processing real clinical data in comparison with a human expert. The study is published in the journal Circulation: Cardiovascular Imaging.
Examination of the heart using magnetic resonance imaging is a technique that allows noninvasively assessing the condition of tissues and departments of this organ, identifying lesions and functional disorders. Due to the accuracy of the data obtained by MRI, they are necessary for making decisions about heart surgery, implant placement, appointment, modification or cancellation of treatment toxic to the heart. Today, an expert describes the captured data manually for decryption. In an uncomplicated case, it takes about 13 minutes per patient. According to the authors of the study, the neural network is able to make at least an accurate analysis in four seconds.
Scientists took the MRI scan data of almost 600 patients and trained a neural network on them. Then, the already trained neural network, expert and trainee were offered the data of new scans of 107 patients from several medical centers. It turned out that with a much higher processing speed, artificial intelligence made mistakes no more often than a human expert. Probably, if the training sample is more accurate, the results of the neural network can become even better, and in the future its diagnoses will be more accurate than human ones.
"Our data set on patients with various heart diseases who have been scanned allows us to demonstrate that most of the measurement errors stem from the human factor. This indicates that automated methods are at least as good as humans, with the potential that will soon "superhuman" change the accuracy of clinical and research measurements," says study author Charlotte Manisty, MD.
Right now, it is impossible to completely switch to analysis using neural networks, but this and subsequent studies are preparing the medical community for change. "Cardiovascular MRI offers unprecedented image quality for assessing the structure and function of the heart; however, modern manual analysis remains basic and outdated. Automated machine learning methods offer the opportunity to change this and radically improve efficiency, and we look forward to further research that could confirm its superiority over human analysis," notes Manisty.
Portal "Eternal youth" http://vechnayamolodost.ru