Towards data-driven predictive temperature control for retinal laser treatments

Laser-induced coagulation is a common treatment of retinal diseases. Manual dosing is prone to error and can lead to extended damage of the neural retina. We elaborated a novel data-driven predictive temperature control (DPC). To identify controller hyperparameters with minimal deviation from the target temperature extensive simulation was carried out, based on 206 measured data sets, each applied pulse energy and induced temperature rise. This resulted in a median and mean deviation from the target temperature of 1.6 °C and 2.6 °C (95 % CI = 2.3 - 3.1 °C), respectively. In addition, we developed a detection algorithm for eye movements (saccades), optimized with Bayesian Optimization that achieved detection sensitivity of 92 % and specificity of 100 %. The experience with Bayesian Optimization shall be used to further optimize the DPC and to evaluate clinical data.