Diagnosis of Central Canal Spinal Stenosis from Lumbar Mid-sagittal MR Images

Central canal spinal stenosis is a significant ailment that impacts the spinal cord of humans. The current automated diagnostic methods for spinal stenosis are unable to diagnose minor abnormalities that may be present in the spinal cord. This paper introduces a novel approach for diagnosing central canal spinal stenosis from T2-weighted lumbar Magnetic Resonance (MR) images captured in the sagittal plane. A segmentation algorithm based on morphology is implemented for segmenting the spinal cord. After segmenting the image, the spinal cord’s skeleton is extracted, and overlapping sub-images are generated. The hybrid of statistical texture features and a novel Spinal Cord Projection Descriptor (SCPD) are extracted for all subimages. Using these features, a k-Nearest Neighbors (k-NN) classifier is trained and a model is constructed. Upon receiving a new test image, the system undergoes these processes to generate features. The features are then utilized by the model to classify the sub-image as either stenosis or non-stenosis. To evaluate the effectiveness of the model, a k-fold validation technique is employed. The constructed model has demonstrated an improved performance in accurately identifying the stenosis region and non-stenosis region, achieving an accuracy rate of 95.19%. The experiments conducted illustrate that the proposed stenosis diagnostic system outperforms existing methods in terms of its effectiveness and accuracy. This study provides promising results for the development of an effective automated diagnostic system for central canal stenosis, which could be of significant benefit to medical professionals and patients. Further research could focus on validating the proposed approach on larger datasets and integrating it into clinical practice.