ART Neural Networks in the Classification of Spinal Pathologies

Spinal diseases are among the significant public health problems and have a negative impact on patients' quality of life. Of these diseases, herniated disc and spondylolisthesis are examples of spinal pathologies that cause severe pain. Currently, in various medical problems related to the diagnosis of diseases, Machine Learning techniques have been used, especially Artificial Neural Networks. From the attributes of the spine such as pelvic incidence angle, pelvic tilt, sacral angulation, pelvic radius, lumbar lordosis angle, and degree of sliding, pattern recognition techniques can be employed to classify herniated disc pathologies and spondylolisthesis. Thus, this paper presents the results obtained by using neural networks of the Adaptive Resonance Theory (ART) family for the classification of spinal pathologies, comparing the results obtained by the different ART networks with those obtained in the literature. Using a neural network in a classification problem has the advantage of robust, stable, fast models that are capable of classification even with little data about the problem. In particular, ART Fast networks are characterized by the ability, even from a few data, to classify the data and, according to the network, will expand from the insertion of new data, improving the chance right. It is also noteworthy that ART extit {Fast} networks perform the classification process faster than traditional ART networks, maintaining the number of hits. The methodology adopted is based on the implementation of ART networks using the Vertebral Column Data Set database, available in the UCI Machine Learning Repository, http://archive.ics.uci.edu/ml/. The results obtained by the network meant satisfactory, obtaining an accuracy of 91.26% for the binary classification problem and 90.97% for the ternary.