A Novel Similarity Measure for Retinal Optical Coherence Tomography Images

Optical coherence tomography (OCT) has become an essential tool for imaging of the retina to perform diagnosis and management of ocular and non-ocular diseases. In this paper, we evaluate the performance of structural similarity (SSIM) and multiscale structural similarity (MS-SSIM) metrics in comparing cross-sectional OCT images of the retina. We will then propose new metrics that only consider areas with a high signal-to-noise ratio in retinal OCT images. Our experimental comparisons demonstrate that, as opposed to SSIM and MS-SSIM measurements, the proposed metrics, called ROI-SSIM and ROI-MS -SSIM, more effectively extract structural information from OCT images and hence, yield similarity values more reflective of human visual perception.