Возможности искусственного интеллекта в измерении оттисков внутриглазного давления по Маклакову

OBJECTIVE: To assess the accuracy of Maklakov intraocular pressure imprints measurement by doctors and artificial intelligence. MATERIALS AND METHODS : Two pairs of tonograms were prepared, obtained by a Maklakov tonometer with a load of 10.0 g. The tonograms were labeled anonymously, using a measuring ruler devised by prof. B.L. Polyak for 4 Maklakov tonometers. In total, 57 ophthalmologists took part in the work. A total of 40 prints were chosen based on there quality. The same prints were photographed by a Xiaomi mi smartphone camera 40 times with a different level of illumination and a different angle of rotation relative to the normal of the lens focal plane. Received photos in jpg format were analyzed by http://ai-tonometry.com algorithms and processed by the Contrast Limited Adaptive Histogram Equalization (CLAHE) algorithm to extract sharper boundaries, and then translated into binary matrices (matrices consisting of “0” and “1”). RESULTS: An imprint with a maximum number of measurements of 40 and a collegially accepted reliable tonometric level of 17 mmHg was measured in the middle range of 16.48±2.7 16.0 (15.0; 17) mmHg by doctors, and 17.0±1.1 17.0 (16.0; 17.0) mmHg by the neural network. At the same time, the range of imprint diameter measurements by the neural network was almost three times smaller, than human measurements. CONCLUSION : The artificial intelligence-based mobile application allows for a high-quality monitoring of intraocular pressure and rejects prints of unsatisfactory quality, which may potentially reduce the number of patients with glaucoma progression.