Automatic Retinogeniculate Visual Pathway Identification Based on Data-driven Fiber Clustering and Anatomical Constrains

The visual system relies heavily on the retinogeniculate visual pathway (RGVP). The identification of RGVP has been accomplished using tractography based on diffusion MRI. Yet, owing to its extremely curved course and complex anatomical setting, RGVP tractography still presents obstacles. The huge false-positive fibers produced by RGVP tractography, which necessitate the labor-intensive hand-drawing of ROIs for fiber filtering, are one of the main obstacles. In order to enable automatic RGVP detection in dMRI tractography, we have proposed a pipeline. First, we created an RGVP atlas based on tractography. Using high-resolution data from 50 cases, multi-fiber unscented Kalman filter tractography was implemented in this study. Then, using a common space created from the 50 tractography cases, we applied data-driven fiber clustering to put nearby fibers with comparable trajectory into one cluster. The RGVP annotation of tractography atlas was done by two qualified anatomists. Second, 50 testing data were used to identify subject-specific RGVP using the RGVP atlas. Also, we developed a deep learning model to assist in screening RGVP clusters and reduce the false positive fibers in subject-specific RGVP. In terms of identification rate, hausdorff distance, and visualization, experimental findings demonstrated that our automatic identification results have perfect colocalization with expert manual identification.