AI models enhance medical diagnostics using routine clinical data

New artificial intelligence models are being developed to enhance medical diagnostics by analyzing routine clinical data.

One system, EpiScreen, utilizes Large Language Models to scan free-text notes within electronic health records to distinguish between epilepsy and psychogenic non-epileptic seizures. This tool aims to facilitate earlier targeted investigations without requiring additional tests. In studies involving the MIMIC-IV database and an external cohort from the University of Minnesota, the model demonstrated high accuracy, with an AUC reaching up to 0.98 in the latter. Research indicates that neurologists using EpiScreen performed up to 14.3 percent better than those without AI assistance.

In cardiology, researchers from Imperial College London presented ECG-CLIP at the European Society of Cardiology congress. This AI model analyzes routine electrocardiograms to predict structural heart diseases. Tested on tens of thousands of patients in the US, the algorithm identified up to 80 percent of individuals with heart failure and 90 percent of those with significant valvular disease. The technology is intended to prioritize patients for ultrasound examinations rather than replace them. Clinical trials are currently underway with the National Health Service in London and Bristol.