Fig. 5: Illustration of a possible application of LAURA beta-forecast.
From: Transformer-based long-term predictor of subthalamic beta activity in Parkinson’s disease
The incorporation of LAURA into the existing aDBS programming workflow would represent a substantial advancement in the field of personalized medicine, with the potential to facilitate the widespread adoption of aDBS therapy as a long-term treatment strategy for PD. For the sake of clarity, the distributions have been presented as minimum–maximum normalized occurrences over beta power, by resampling and taking the average of each successive pair of bins. At the first visit (top-left, grey panel), the neurologist defines the optimal parameters for aDBS based on the clinical evaluation and previous subthalamic recordings. This entails defining the percentiles of the daily distribution below βmin and above βmax. These two thresholds allow to define the amount of time the patient is at Amin, Amax, or actually linearly adapted. LAURA receives as input a sequence of N* (=2 for patient NWK1; day −2 and day −1, top row in red) observed daily distributions and provides as output the predicted subthalamic beta power distributions over 6 days in advance (bottom row in green), together with a reliable estimate of their tails (i.e., the percentiles of the distribution below βmin and above βmax). This allows alarms to be set whenever the tails exceed a safe range (e.g., 10%) around thresholds for aDBS reprogramming and paves the way for remote monitoring strategies and the implementation of new algorithm for personalized auto-tuning devices.
