부산대학교 도서관

Visual Field (VF), a measurement of retinal function, is one of the most important references in modern glaucoma diagnosis and treatment. While VF is a typical physical-psychological measurement, the results are often imprecise. How to improve measurement accuracy is always one of the most significant issues in clinical practice. This study proposes a VF2VF training method based on the assumption that the pattern of multiple VF measurements in a short period does not change, and designs a neural network to transform the VF measurements. We analyze the VFs after transformation through experiments. Experiments show that the training method of VF2VF can significantly improve the precision while maintaining the accuracy of original VFs. Besides, the VFs after transformation achieve a significant performance improvement in downstream deterioration detection tasks. When the false positive rate is 5%, the Hit Rate increases by 20%. And the transformed VFs can give a warning 1.8 times ahead of the original VFs.