A method for measuring intraocular pressure using artificial intelligence technology and fixed-force applanation tonometry

Abstract

Purpose. To estimate the accuracy of IOP measurement using artificial intelligence (AI) technologies and applanation tonometry with fixed strength. Material and methods. 290 patients (576 eyes) underwent applanation tonometry according to Maklakov with tonometer weights of 5, 10, and 15 g using a modified elastotonometry technique followed by an analysis of impression quality and diameter measurements by three independent ophthalmologist experts. The prints were then fed into a neural network to check the repeatability and reproducibility of the measurements. Results. The comparison of the diameters of the Maklakov tonometer prints determined by AI based on the neural network with the measurements data provided by three experts showed that neural network underestimates the measurement results by an average of 0.27 (-3.81; 4.35) mm Hg. At the same time, the intraclass correlation coefficient for all prints was 98.3%. The accuracy of diameter measurements of prints by neural network differs for tonometers of different weights, e.g. for a 5 g tonometer the difference was 0.06 (-3.38; 3.49) mm Hg, for 10 g and 15 g tonometers was 0.14 (-3.8; 3.51) and 0.95 (-3.84; 5.74) mm Hg, respectively. Conclusion. High accuracy and reproducibility of the measurements by the neural network, was shown to surpass the reproducibility of human-implemented measurements.

Keywords

• глаукома
• аппланационная тонометрия
• внутриглазное давление
• офтальмотонометрия
• искусственный интеллект
• тонометрия с фиксированной силой