Recently, in vivo visualization of the cochlea and the smaller structures inside of it has been achieved by optical coherence tomography (OCT). This makes it possible to use OCT imaging for diagnosis of diseases such as Meniere's disease through measuring the degree of endolymphatic hydrops. To this end, we present a novel method for 3D segmentation of these cochlear OCT images that is based on superpixels and diffusion maps. The method takes as input grayscale volumetric OCT images and outputs a binary image with the segmented cochlea. We show that the proposed method is suitable for segmenting the data for visualization as well as for preprocessing the data for future automated grading of endolymphatic hydrops.