Abstract:
The world is confronted with the issue of backbone pain, which is becoming critical and a serious
clinical issue, and spine diseases are becoming a growing social and medical problem worldwide.
Lumbar spondylosis is a type of disk disease that occurs primarily in the lumbar region, owing to
an unfavorable ratio of the mechanical load to the size of the intervertebral discs. Lumbar
spondylosis is one of the leading causes of morbidity and disability, and its prevalence is
increasing as the world's population ages. The preferred imaging modality for identifying the
causes of complicated lower back pain is MRI.
Numerous studies on the diagnosis of lumbar spondylosis have been conducted in the past, but
they were not sufficiently thorough due to a lack of data, preprocessing techniques, manual
segmentation, poor feature learning techniques, evaluation metrics, and classification techniques
that are effective at picking out the ideal characteristics that may reflect the most helpful contents
of images. Our study uses TensorFlow and the karas API to build a lumbar spondylosis
classification model based on an ensemble of convolutional neural networks feature extraction.
The study used 11158 T1 and T2 MRI images with a size of 64X64 PNG format that determines
the type of spine abnormalities type (disc bulge, disc desiccation, nerve root compression, and
healthy) both axial and sagittal scan modes taken under the supervision of a chief radiologist from
the university of Gondar specialized hospital. We extracted the most important part of an image
using the center crop, then used mean, median, Gaussian filter, and AHE techniques to remove
some noise and kmeans, Otsu, and threshold segmentation algorithms to segment the image.
The dataset was divided into 80:20 ratios and fed to ensemble CN
