Construction And Application Of The U-Net Network Model In Pupillary Light Reflex Experimental In Mice

Background: The pupil is an extremely important part of visual regulation,regulating the light entering the eye through changes in its size to achieve clarity of objects falling on the retina.The pupillary light reflex(PLR)is a fundamental reflex phenomenon closely related to the visual system and the autonomic nervous system,in which the pupil reflexively narrows when the eye is stimulated by external light to reduce the amount of light entering the retina.Diseases are often used as important indicators for diagnosis and evaluation.However,as the data from PLR experiments are mined in depth,data processing becomes more and more laborious and time-consuming,and ordinary manual testing is no longer sufficient for data processing.Therefore,there is an urgent need to combine U-Net graphical prediction with PLR to invent an automatic pupil recognition system.Aims: Firstly,to establish a pupil detection system with rapid and accurate detection using the U-Net network in neural convolutional networks.Secondly,to analyse the changes of PLR in wild-type mice C57BL/6J and retinal degeneration model mice rd10 at different ages and different light intensities.Finally,the U-Net model was used to process the pupil images in the PLR experiments for long-term monitoring of the treatment effects in rd10 mice treated with gene replacement therapy.Methods:(1)The U-Net model in convolutional neural networks was used to train and build a pupil detection system.(2)The results of pupil-to-light reflex experiments in C57BL/6J mice as well as rd10 mice under different light and age were analysed using two different outcome analysis methods,U-Net prediction and manual detection.(3)To observe the long-term therapeutic effect of gene replacement therapy,the pupil contraction of treated rd10 mice was monitored by PLR assay at 7 light intensities of5 lux,20lux,50 lux,100lux,200 lux,500lux and 1000 lux in a cycle of about 20 days.UNet was used instead of manual monitoring to analyse the long-term effects of gene therapy.Conclusions:(1)The U-Net network model was more accurate in predicting mouse pupil size under the training of more than 300 sets of pupil photographs of C57BL/6J and rd10 mice in PLR experiments,and the average accuracy of detection reached96.77%.(2)The pupil detection system based on the U-Net model was applied to the PLR experiment for validation.In the comparison of the pupil maximum constriction amplitude,the average error was only 0.46% and the accuracy was high in both C57BL/6J mice and rd10 mice.(3)Analysis of the effect of different conditions on the pupillary reflexes of C57BL/6J mice using the U-Net network revealed that the pupillary reflex phenomenon and ability in C57BL/6J mice was not related to the age of the mice,and its change only increased with the increase of light intensity,reaching the maximum constriction amplitude after 500 lux,which was about 82%.rd10 mice had a maximum pupillary constriction amplitude of P16 The amplitude of pupil contraction was similar to that of C57,and the pupillary reflex to light was almost completely lost at P90.(4)Gene replacement therapy with AAV-pde6 b significantly reduced the rate of retinal degeneration in rd10 mice,and U-Net analysis showed that this treatment delayed the time of massive optic rod cell death from around P60 to between P122 and P137,and that the retinal photoreceptors of rd10 mice remained stronger than those of untreated mice until P183.