Zhongshan Eye Center is the world's first artificial intelligence assessment technology for infant visual function

Zhongshan Eye Center is the world's first artificial intelligence assessment technology for infant visual function

Due to the difficulty of objectively assessing the visual function of infants and young children, more than 20 million infants and young children with visual impairment in the world cannot be detected in time and become blind for life, causing a heavy social burden. Professor Liu Yizhi and Professor Lin Haotian of Zhongshan Eye Center of Sun Yat-sen University discovered the differences in the behavior patterns of infants and young children with normal and visual impairment, and used deep learning technology to establish the world's first intelligent visual function assessment system for infants and young children based on behavior patterns, which is used to objectively screen the visual function of infants and young children and detect visual impairment in infants and young children before language in time. The latest research finding, "Discrimination of the behavioural dynamics of visually impaired infants via deep learning", was published in Nature Biomedical Engineering, a sub-journal of the journal Nature, on October 21, 2019. 鈻睺he homepage of the article published online by Nature Biomedical Engineering鈻睴n October 22, Professor Liu Yizhi, Professor Lin Haotian and their team released this blockbuster research result to the media. For the first time, the quantitative relationship between visual impairment and behavior patterns is clarified, and the precise coordination of perception and behavior is the basis for biological survival and evolution. Vision is the most important human perception, and previous studies have shown that there is a certain correspondence between visual and behavioral phenotypes. However, how vision loss affects changes in individual behavior patterns remains largely unknown. In this study, we analyzed the behavioral phenotypic video big data of 4196 infants and young children, quantitatively compared the frequency and severity of 13 behavioral characteristics in 4 categories of different visual function groups, and clarified for the first time the quantitative relationship between 11 landmark medical behavioral signs such as strabismus, nystagmus, and compensated head position and visual impairment in infants and young children. 鈻睺he research process of the quantitative relationship between visual impairment and behavioral patterns innovatively applies the medical artificial intelligence algorithm framework with the temporal segmentation network as the core, and this research innovatively uses the temporal segment network to learn from the main and establish a characteristic model of behavioral phenotypes at the video level. The algorithm combines the strategy of sparse sampling and video-level supervision to achieve efficient fitting of dynamic video datasets. To put it simply, the algorithm adopts a sparse sampling scheme based on segmentation structure to extract short clips from the original video sequence, and then completes the prediction and inference of the video level from the aggregation information of the sampled segments. The auxiliary optical flow network will be integrated through the consistency segmentation function to achieve the effect of score fusion of different fragments to generate the final classification probability. 鈻睧stablishment of an infant intelligent visual function assessment system using temporal segmentation network self-learning phenotypic characteristicsThis study provides the possibility of establishing a highly accurate and specific infant intelligent visual function assessment system. The results of the study showed that the intelligent assessment system performed satisfactorily in detecting mild and severe visual impairment and in diagnosing the etiology of infants and young children by assessing the visual function of infants and young children through video recording of behavioral patterns. Compared to traditional testing methods, the system requires less technical support and infant collaboration, and is more feasible and accurate. In addition, the system can be used as a technical support for clinical research on visual development, which is of great significance for further exploring and clarifying the laws of visual development in infants and young children. 鈻睞rtificial intelligence screening model is used to clarify the cause of visual defects in infants and young children, with excellent sensitivity and specificityResearch Background IntroductionZhongshan Eye Center of Sun Yat-sen University has been committed to the research and development and innovation of medical technology, leading the development of ophthalmology services. In May 2018, the first national new clinical specialty of medical artificial intelligence was established, positioning the three major tasks of artificial intelligence diagnosis and treatment platform research and development, clinical trials, and application service promotion. Under the leadership of Prof. Liu Yizhi and Prof. Lin Haotian, the department has completed the construction of the ophthalmology big data resource pool (Science 2015), developed new cataract therapies and successfully applied them to clinical practice (Nature 2016), and created the world's first artificial intelligence cataract diagnosis and treatment cloud platform (Nat. Biomed. Eng. 2017, cover paper) and the intelligent prediction system for adolescents with myopia (PLOS Medicine 2018), opened the world's first ophthalmic artificial intelligence robot clinic, and completed the world's first artificial intelligence multi-center clinical controlled trial study (EClinicalMedicine 2019, cover paper) with a number of ophthalmology units. With the support of major scientific research projects such as the National Key R&D Program, the Guangdong-Hong Kong-Macao Greater Bay Area Medical Technology Development Key Project and the Guangdong Provincial Key Area R&D Program, the specialty has reached the international advanced level in the field of artificial intelligence medicine.