Comparison between K-means and Self-Organizing Maps algorithms used for diagnosis spinal column patients

Back pain is a problem that affects at least 90% of the human population during their lifetimes, and it is one of the main causes of absence from work. Despite the large number of medical cases, there is insufficient research in this field, due in part to the necessity of administering multiple medical exams to patients with vertebral problems for accurate diagnosis. In addition, the cause of back pain may be benign, and heal without treatment. In this study, two artificial intelligence algorithms were employed to classify patients with spinal problems. The algorithms used were K-means and Self Organizing Maps (SOM). With these techniques, two models were obtained that provided a generalization error of less than 10%. The models were compared based on metrics that enable the measurement of classifier performance, including the sensitivity, specificity, precision, and negative predictive value (NPV), as well as Cohen's Kappa index to evaluate concordance. It was found that the model trained with SOM outperformed the model trained with K-means, with improved detection of patients having vertebral problems. Additionally, it was found that the SOM and K-means models yielded similar precision as compared with models obtained with different algorithms reported elsewhere. The values yielded were in agreement with those of expert orthopedic physicians. Keywords: Machine learning, K-means, Self-Organizing Maps, Unsupervised learning, Clustering, Kappa coefficient, Confusion matrix