Generation of synthetic aqueous flare data using deep learning models

Synthetic data provides a compelling solution for addressing issues regarding confidentiality, streamlining information usage agreements, proposing procedures, obtaining regulatory approvals, and reducing expenditures. Deep learning (DL) models excel at creating time series data due to their exceptional capacity to reproduce the patterns in the data. This work utilized long short-term memory (LSTM), bidirectional long short-term memory (BiLSTM), and gated recurrent unit (GRU) models to generate synthetic aqueous flare data. Aqueous flare is a measure of the intensity of light scattering in the anterior chamber of the eye, indicating the level of inflammation. The obtained data is a time series that was measured in an experimental environment, which includes the measurement of blink artifacts. The performance evaluation of each of the three models was undertaken in this study by altering the parameters of the deep learning models. The primary objective of the project is to determine the optimal model for producing synthetic aqueous flare data. The results demonstrate that the GRU model has shown exceptional performance in synthesizing aqueous flare data.