In order to solve the problem of the influence of different optical coherence tomography (OCT) image preprocessing methods on scattering coefficient calculation in dermis, and to improve the accuracy of noninvasive blood glucose detection, an optimization method of early OCT image data processing was proposed. The 3-D images were first processed by skin surface alignment, 3-D reconstruction and 1-D average, and then the background noise and data normalization effect on the accuracy of blood glucose prediction were analyzed by clinical experiments. The results show that the prediction error decreases by 18.31% compared with that before pretreatment. This study has important reference value for improving the accuracy of optical noninvasive blood glucose detection using OCT.