Automatic localization and classification of intervertebral disc herniation using hybrid classifier

Abstract

Lumbar disc herniation detection is challenging due to numerous problems, including a complex background, and small joints. Deep learning-based methods have set new benchmarks for much computer vision and pattern recognition research throughout the last five years. In this proposed work, a novel method for the localization of horizontal and vertical projection of Intervertebral Discs (IVDs) in lumbar spine MRI images based on YOLOv2-IVD (You only look once v2-Intervertebral Disc) is suggested, which includes Enhanced Visual Geometry Group 16 (EVGG16) as the YOLOv2 model's backbone. In this framework, features are taken from the ReLU 11, which is the output layer of EVGG16 modeland fed into the YOLOv2 model. Following that, the exact intervertebral discs are detected by finding the intersection of the IVDs localized horizontal and vertical projections, from which IVDs features are learned using Enhanced AlexNet (EAlexNet). It includes 8 weight layers and kernels 5×5,3×3 for classifying IVDs as normal or herniated, a process also known as disc herniation classification. Finally, the severity of the herniation, such as prolapse (mild) and extrusion (moderate), is determined by estimating the length of the intervertebral herniation disc, which has been achieved using basic image processing techniques such as thresholding and morphological segmentation methods. The proposed method is trained and validated using a real-world spine MRI image dataset obtained from a hospital. Also the results reveal that the proposed approach, attains a high accuracy of 93.59% for herniation detection, comfortably outperformed other state-of-the-art methodologies.