In common retinal disorders like Epi-Retinal Mem-brane (ERM), accurate prediction of postoperative visual outcome is essential to facilitate early surgical intervention and appropriate treatment. However, predicting postoperative visual outcomes for individual patients is difficult for medical doctors in the current clinical setting. From the clinical background viewpoint, we have been researching quantitative postoperative prediction by machine learning using OCT images. Currently, during the process of extracting features from given OCT images, annotation is performed by medical doctors. However, this work requires much time and is a burden for medical doctors. Therefore, this study discussed the effectiveness of features calculated from the fovea to alleviate medical doctors’ burden for annotation. In this paper, feature extraction was performed from B-scan OCT images of ERM patients annotated with fovea pit. The proportions of high-intensity values within the parafovea, fovea, and Foveal Avascular Zone (FAZ) square area were calculated. By using these features, postoperative visual acuity was predicted, and the results were evaluated. The results showed that prediction using features from the fovea pit in the macula was accurate compared to the conventional method. These results suggested that the features obtained from the fovea pit may be significant for prediction.