Comprehensive visualization of the mouse spinal cord using serial optical coherence microscopy and deep learning techniques

Abstract

The morphological observation of the central nervous system (CNS) of mouse models is crucial for identifying the function of the CNS and understanding the neuronal diseases. Although the spinal cord (SC) is the major communication pathway between the brain and the peripheral nervous system, there is lack of effort to visualize and quantify the SC compared to actively investigated the case of the brain. There are also inherent challenges for optical imaging modalities to monitor the volumetric SC, because its long and tubular structure as well as the location surrounded by three layers of meninges. In addition, white matter consists of highly scattered lipids which restrict the light penetration. Therefore, new imaging modality and protocols are essentially required to provide comprehensive visualization of SC. In this study, we suggest a novel imaging approach which enables us to offer high throughput quantitative information as well as native 3D context. Our technique is based on optical coherence microscopy (OCM) and deep learning techniques. OCM utilizes the endogenous back-scattering signal, and it does not require chemical labeling, staining, or external contrast agents. OCM is also suitable for deep tissue imaging because OCM uses near-infrared light that reaches deeper into biological tissues than visible light can reach. In order to collect volumetric and informative anatomy, we built a unique imaging protocol with four steps. First, SC was embedded in agarose gel, sequentially sectioned with 50μm thickness using a vibratome, and imaged by OCM. Second, the process of imaging and sectioning is repeated until the thoracic part of the spinal cord image set is completed. It was then trained by a convolutional neural network (CNN) which is targeted to distinguish the region of the gray and white matters in SC. The performance of the trained network was evaluated and adjusted at volumetric data which was not included in the training dataset. Finally, the virtual segmentation algorithm was completed which has the capability to connect and visualize the connectivity of long SC. In preliminary research, a large field of SC anatomy was successfully constructed through our new imaging approach while delineating a detailed spinal nerve. We also quantitatively concluded the distribution of white and gray matters of the SC in a lengthy direction. Based on experimental results, we confirmed that the recovery process after partial spinal cord injury or multiple sclerosis disability could be monitored with comprehensive means.