Presagen has developed a new algorithm that cleans poor-quality data automatically to improve artificial intelligence (AI) scalability.
Data used to develop and train AI algorithms can sometimes be of poor quality due to clinical subjectivity, uncertainty, and even adversarial attacks, according to Presagen, and sometimes event experts cannot detect data errors.
Poor-quality data can affect the training stability and performance of AI algorithms, and Presagen's UDC technology works by automatically detecting poor-quality data, according to the company.
UDC can be applied for imaging issues, specifically detecting pneumonia on chest x-rays. UDC reliably detected poor-quality data on its own, improving AI accuracy in some cases by more than 20%, Presagen said. The algorithm also cleans data that is used to validate AI accuracy.
