Lumbar Spinal Stenosis Detection from Sagittal and Axial MR Images using Hybrid of Deep Kronecker Network and SpinalNet.

Purpose: The main purpose of this research is to develop an automated and accurate method for detecting Lumbar Spinal Stenosis (LSS) using both sagittal and axial MRI images. The study addresses the challenge of differentiating LSS from similar conditions such as herniated disks, aiming to facilitate quick diagnosis with less dependence on expert interpretation.

Method: This research proposes a Hybrid Deep Kronecker Spinal Network (DKN_Spinal) for LSS detection using sagittal and axial Magnetic Resonance Imaging (MRI) images. Here, the input sagittal and axial images are fed for the pre-processing, which is done by bias field correction. Then, the stenosis region segmentation is carried out by Fuzzy Local Information C-Means (FLICM) method. After that, the LSS is detected for both the segmented images by DKN_Spinal, which is the combination of Deep Kronecker Network (DKN) and SpinalNet. Furthermore, the classification of lumbar spine conditions in sagittal and axial images is performed, categorizing them as mild, moderate, or severe. Finally, the majority voting is carried out for both the categorized phases, which is then classified as mild, moderate, or severe based on the training loss.

Results: The proposed DKN_Spinal model demonstrated superior performance, achieving an accuracy of 92.1%, True Positive Rate (TPR) of 92.2%, and True Negative Rate (TNR) of 92.9%.

Conclusion: The proposed method achieves high diagnostic accuracy and effectively classifies spinal conditions into mild, moderate, or severe, providing detailed insights that support appropriate treatment planning and reduce the need for extensive expert involvement.