High-resolution virtual biopsy: non-invasive pathology using deep learning and optical coherence tomography

Virtual biopsy enables non-invasive diagnosis through machine-learning analysis of high-resolution images. But difficulties in generating accurate co-registered training sets and the resolution/field-of-view (FOV) tradeoff have hindered clinical applications. We present a method enabling reliable and accurate co-registration of 3D cellular-resolution OCT of fresh human skin with downstream histology. Orientation data is encoded across the sample by photobleaching a fiduciary grid pattern into fluorescent gel encasing the tissue. These markers persist in histology sections, permitting accurate co-registration to the 3D volume within ~20µm, and enabling cellular-resolution imaging with a cm-level FOV by laterally tiling OCT volumes, crucial steps toward in-vivo high-resolution Virtual Biopsy.