Arana Estanislao
This thesis is dedicated to assess the accuracy of logistic regression (LR) and artificial neural networks (ANN) in the diagnosis of calvarial lesions using computed tomography (CT). The importance of the different features needed for the diagnosis in both models is also analyzed. The models were developed using patients with calvarial lesions as the only known disease were enrolled. All patients were studied with plain films and CT. Other imaging thecniques were used when available. The clinical and CT data were used for developing LR and ANN models. Both models were tested with the jacknife (leave-one-out) method. The best ANNs were obtained varying iterations and hidden neurons by selecting the one with higher area under the receiver operating characteristic curve (ROC). The final results of each model were compared by means of area under ROC curves. There was no statistically significant difference between LR and ANN in differentiating benign and malingnant lesions. In characterizing every histologic diagnoses, ANN was statistically superior to LR (p<0.001). ANNs were demonstrated adequate to diagnose the most common lesions in the cranial vault and to help the radiologist with misleading and infrequent appearances. ANN discover hidden interactions among variables that are missed in the statistical analysis.
