|
Abstract:
|
Deep Learning Artificial Intelligence is seeing accelerating growth for providing diagnostic information and physiological monitoring information to healthcare providers, and also for providing information to laypersons. One example area where FDA is seeing increasing interest is medical imaging where convolutional neural networks are being used and proposed for classification of disease conditions. Other medical areas where FDA is seeing increased interest is analysis and classification of time series data, such as from electrocardiography (ECG) and electroencephalography (EEG) devices. While challenges exist in performance validation and statistical analysis of AI-enabled medical devices, certain existing methodologies can still apply effectively. However, further research in statistical analysis methodologies should be able to help improve accuracy and patient outcomes. Promising areas of research include statistical methods to assess combination of physician assessments with AI-enabled devices and combination of two or more AI devices, to produce increased combined sensitivities, specificities, positive and negative predictive values, etc.
|
