Data science is the study of data analysis, but this situation has been changing dramatically recently. Artificial intelligence (AI), fifth-generation (5G) communication networks, and the internet to things (IoT) are being implemented in society and are changing society at a rapid pace. In medicine, this change is called the digital medical revolution, and it is spreading worldwide. We will introduce our research on this change.
I. Potential of AI in image analysis
In ophthalmology, images are often used for medical treatment and research, but have insufficient objectivity and quantitativeness. Therefore, we conducted research on objective and quantitative image analysis using AI. We attempted to stratify en face optical coherence tomography (OCT) image of the choroid. We developed an algorithm for the determination of en face OCT image layers of the choroid using AI with support vector machine (SVM). An SVM model was built using each attribute in the training data, and the coefficient of determination for each layer of the choroid was calculated using the validation data. The results showed that the coefficient of determination was 0.985 on average. This indicates that images that could only be distinguished qualitatively can be classified objectively and reproducibly by AI. We classified the running pattern of choroidal vessels in en face images obtained by this method. We acquired en face images from eye with central serous chorioretinopathy (CSC) and calculated the vessel area, vessel length, and mean vessel diameter. The symmetry index was calculated as the percentage of vessels running in the symmetrical direction. The result showed that blood vessels were dilated and ran asymmetrically up and down in CSC eyes. Since this structural difference was observed not only in the diseased eye but also in the fellow eye, we confirmed the existence of a congenital predisposition in CSC. There are no thick blood vessels in Haller layer in the choroid just below the macula in normal eyes, reducing the influence of vasodilation on the macula. In CSC eyes, as the Haller blood vessels run asymmetrically, Haller vessel dilation excludes the inner layer of the choroid below the macula, disrupting the blood-retinal barriers and showing accumulation of exudates under the retina.
II. The development of new research themes using AI
Although AI has proven that gender determination using fundus images is possible, the proving process is unclear. To elucidate the process, we collected quantified accumulated data related to fundus photographs and investigated whether it is possible to predict gender from these data alone. As a result, we could predict the gender of the fundus with approximately 80% accuracy. The female fundus was greener than the male fundus and it was characterized by the superior temporal artery running close to the macula. The next study, which investigated what these findings reflect, revealed that approximately 60% of the characteristics of gender differences were expressed at birth, and the percentage increased to 80% after the secondary sex characteristics. The fact that AI discovered phenomena and facts that humans had never thought of, and that humans studied them, has great significance in the history of ophthalmology. In the future, AI may become the main source of research themes.
III. Importance of big data
Big data are at the heart of data science in future ophthalmology. The Japanese Retina Vitreous Society (JRVS) registered all retinal detachment surgeries performed in JRVS board member institutions in 2016, and 3446 cases were accumulated (J-RD Registry). Various analyses were conducted, and basic data on retinal detachment in Japan as of 2016 were completed, including the tendency of surgeons to choose pars plana vitrectomy and scleral buckling, changes in visual acuity, and differences in the success rate of surgery depending on the detachment site. We further analyzed these data using propensity score matching, and we obtained evidence indicating that silicone tamponade causes a more severe retinal damage than gas tamponade, and that internal subretinal fluid drainage will promote postoperative preretinal membrane formation. Additionally, many other findings, such as gender differences in treatment, are still being obtained. Big data are important to proceed with the future eye care correctly. However, Japan lags behind in this field, and should establish it early.
Nippon Ganka Gakkai Zasshi (J Jpn Ophthalmol Soc) 126: 221-253,2022.