Segmentation and classification of intervertebral disc using capsule stacked autoencoder

Abstract

Accurate location and segmentation of IVD (intervertebral disc) are important for detecting spinal cord diseases. Despite tremendous advances in medical imaging, IVD localization and segmentation remains a hot research topic. The spine is exposed to various kinds of compression forces, which seriously causes the function of the disc and causes disorders like disc bulging and herniation. This work uses CAD (computer aided design) for IVD localization and segmentation using axial MRI images of the spine. DL (deep learning) models ensure speed and accuracy in diagnosing the process and classifying the spine problem. Hence, this work employs a DL based IVD localization and segmentation. Initially, the image is resized and enhanced by AHE (Adaptive Histogram equalization). Then, the Mask-Region-based Convolutional Neural Network (Mask R-CNN) is used to localize IVD. After localizing, the SegNet (semantic segmentation network) model efficiently carries out disc segmentation. Then the weights of SegNet are optimally fine-tuned by the adaptive rain optimization algorithm (AROA). Finally, the DL model capsule stacked autoencoder (CSAE) is used for classifying the types of diseases in IVD. The overall implementation in the Python platform and analysis in the benchmark dataset Lumbar Spine MRI achieved better accuracy and precision of 98.5% and 97.8%, respectively.