BlinkNet-Q: A real-time pupil rate measurement software based on deep learning that identifies incom
An international medical study has yielded new findings that have important reference value for disease prevention codes. The development and validation of recent real-time quantitative computing of blink rates and incomplete blinking software based on the dual-depth learning model DLM-1 and DLM-2 (BlinkNet-Q) has resulted in the development and validation of two DLMs: DLM-1 distinguishes between open-eye and fully-blinking, while DLM-2 distinguishes completely from incomplete blinking software. This proof-of-concept technology validation study includes the development of separate models and an independent test cohort.
The development of the cohort consisted of 15 participants (*30 eyes) and produced 29,53 trained and internally verified data sets that were marked with a lens, which were split into participant-level data sets. The test cohort consisted of 39 participants (*39 eyes) and was used for the final evaluation of BlinkNet-Q. The evaluation indicators used classification indicators ((C) ), regression based on indicators ((R) ), intrinsic coefficients ((ICC) and Bland-Altman analysis.
The AUC values obtained by the DLM-1 system were 0. 989 between 95% confidence interval [CI], 0. 932-0.
993) and the AUC values obtained by the DLM-2 system were 0. 884 between 95% confidence interval [CI], 0. 854-0.
963). The results showed that BlinkNet-Q and manual counting blink rates were between . ICC=0.
973; R^2=0. 948; mean absolute error [MAE] = 1. 262 blinks/minute) and incomplete-blink ratio (ICC = 0.
937; R^2 = 0. 876; MAE = 0. 059) are well consistent and have very small average deviations under standardised recording conditions.
BlinkNet-Q demonstrates the technical feasibility of automated blinking and incomplete blinking under standardised recording conditions, supporting its potential as a prototype of camera-based blinking assessment software. Clinical implementation may need to be carried out prior to external verification of different clinical populations. BlinkNet-Q may contribute to the future development of accessible, camera-based eyelid assessment tools for eye surface assessment.
It should be noted that this is based on the results of international public research and is for reference in the HealthCop and does not constitute a clinical recommendation. If you have any health questions, consult professional medical professionals.
