Shannon entropy and fuzzy C-means weighting for AI-based diagnosis of vertebral column diseases

• Published in: Journal of ambient intelligence and humanized computing Vol. 11; no. 6; pp. 2557 - 2566
• Main Authors: Alafeef, Maha, Fraiwan, Mohammad, Alkhalaf, Hussain, Audat, Ziad
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.06.2020; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial Intelligence; Artificial neural networks; Biomechanics; Classification; Clustering; Computational Intelligence; Computer aided decision processes; Datasets; Decision making; Decision trees; Diagnosis; Disease; Engineering; Entropy; Entropy (Information theory); Hospitals; Machine learning; Magnetic resonance imaging; Neural networks; Original Research; Orthopedics; Pelvis; Robotics and Automation; Spinal cord; Surgeons; User Interfaces and Human Computer Interaction; Vertebrae; Weighting; X-rays; Shannon entropy; Spondylolisthesis; Fuzzy C-means; Vertebral column diseases; Herniated disk
• ISSN: 1868-5137; 1868-5145
• DOI: 10.1007/s12652-019-01312-3

Degenerative vertebral column diseases are becoming increasingly common and computer-aided decision-making and diagnosis systems are gaining popularity. In this paper, we propose a machine learning decision-making model based on noninvasive panoramic radiographs to tackle the problem of automated diagnosis of two common vertebral column diseases; disc prolapse and spondylolisthesis. We collected raw data from real X-ray images of 422 subjects (i.e., 201 disc prolapse, 111 spondylolisthesis, and 110 healthy). We used five biomechanical parameters as input to the model representing the pelvic incidence, pelvic tilt, lumbar lordosis angle, sacral slope, and degree spondylolisthesis. To obtain more meaningful features, we preprocessed each vertebral column dataset by weighting every vertebral feature using a set of weights computed based on Shannon entropy and the fuzzy C-means clustering algorithm. Then, the new weighted set of features was fed to an artificial neural network classifier. Our proposed method was able to classify the subjects into three classes with 99.5% overall accuracy. This reflects a strong ability to predict the patient vertebral column dysfunction using the biomechanical attributes and with an accuracy satisfying clinical requirements. This approach represents a feasible system that facilitates the diagnosis of vertebral column disorders. It can help the physician to take the correct decision very early, which will prevent the development of the pathology into a chronic level and reduce the need for surgical treatment.