Discrimination of Osteopathic Patients using Logistic Regression Model

Osteoporosis is an illness characterized by a bone fragility, which exposes a person to a fracture risk. The purpose of this work is to assess osteoporosis using different attributes of radiographic image texture, and clinical parameters of two populations of patients, composed of 70 healthy subjects, and 70 osteoporotic cases. Different configurations of eight parameters were involved to improve the ability of the discrimination between the groups. Four models were trained using a Logistic regression, and validated by the so called 10-fold cross-validation” technique. A combination of the clinical and the textural parameters was proposed to achieve a high performance of the classification. The statistical analysis was done using the “non-parametric” statistical Wilcoxon test, confusion matrix, and ROC curves. The results obtained reveal an accuracy of S6.4% and an Area Under ROC (AUC) of 96.3%, for the proposed configuration demonstrating the high performance for osteoporosis assessment.