Estimation Method of Visual Prognosis in Branch Retinal Vein Occlusion with Macular Edema

Branch Retinal Vein Occlusion (BRVO) with Macular Edema (ME) is a common retinal vascular disease and needs a remedy for an extended period. There is a high possibility that patients lose motivation for treatments. Therefore, visual prognostic evaluation is essential to reduce anxiety and increase motivation. However, in present clinical practice, it is challenging for ophthalmologists to predict the prognosis individually. On this clinical background, this study aims to establish a method for estimating visual prognosis in BRVO with ME. In this paper, we attempted binary classification using clinical data and Machine Learning techniques, in which patients are classified into the good or poor visual prognostic group. Specifically, we designed the classifier to predict the good prognostic group perfectly. That is because the most important thing is improving the patient’s motivation. In this experiment, we targeted 66 patients who were followed up for 12 months at Mie University Hospital. Logistic Regression was employed as a learning algorithm. After extracting the features from clinical data, we performed 5-fold stratified cross-validation. As a result, we achieved a mean precision for the good group of 0.95 and 0.80 for the poor group. In addition, the mean precision for the poor group in training data was 0.86 when the classifier satisfied the precision for the good group in training data of 1.0. This result suggests that the performance for unknown data does not significantly decline relative to the performance on known data. Furthermore, obtaining this performance using a linear classifier has an additional advantage because the prediction process is interpretable. We expect that this method would provide a better treatment experience if patients receive the models’ performance, the number of training samples, and the prediction process, in addition to the diagnosis.