Pipelined Multilayer AI-Based Point-of-Care Model for Diagnosis of Spinal Cord Disorders in Big Clinical Data

---

Abstract

The disorders in the spine reduce the quality of human life therefore large clinical data with scanned spinal cord images can be processed by AI-based point-of-care services to quickly and accurately diagnose spinal cord problems such as lumbar spinal stenosis, spinal deformities and spinal osteoarthritis. Vertebrae localization and segmentation are essential in the accurate diagnosis of spinal cord disorders. However, the existing labelling and localization process for scanned images of the spinal cord is not suitable for large clinical data corroboration from numerous patients and has erroneous label findings due to missed vertebrae that were only partially visible on the image, similar forms of vertebrae in cervical, thoracic, and lumbar regions, as well as labelling process, failed to detect sacrum. Also, localization error occurs especially in spinal cord images with anatomical abnormalities such as additional and transitional lumbar vertebrae, it is difficult to properly locate vertebrae in the mid-thoracic area. So, there is a need to develop a novel AI-based point-of-care model for large clinical scanned spine images to effectively diagnose various spinal cord disorders at an early stage to provide timely treatments with accurate labelling and segmentation of spinal cord components. The proposed AI-based point-care-model uses Pipelined labelling with level count Circular Localization, and then Feature Transformer based Classification which effectively diagnoses various spinal cord disorders with its sublevels at an early stage with accurate labelling, segmentation and feature extraction on the localized spinal cord components.

---

Keywords

Deep learning, point of care, AI, Hough line transforms filtering