Inteligência artificial aplicada ao diagnóstico radiográfico de displasia coxofemoral em cães

Abstract

Radiographic imaging is the main diagnostic tool for canine hip dysplasia, through which the degree of the disease is classified and signs of joint laxity, secondary degenerative joint disease and other alterations inherent to the condition are evidenced. However, interpreting these images is a challenging and error-prone task due to variations in breed and sizes, as well as radiographic techniques and positioning. Technologies that combine machine learning, computer vision and medical imaging facilitate disease diagnosis and provide a second opinion for professionals. This study aims to determine the accuracy of a convolutional neural network in detecting normal and abnormal radiographic patterns in hip joints and correctly classifying them among dogs with and without hip dysplasia. The results significantly contributed to creating a groundbreaking national radiographic database aligned with international standards for the classification of canine hip dysplasia. Implementing artificial intelligence tools proved promising, with an average accuracy of 92%, sensitivity of 91%, and specificity of 92%. The ROC curve and AUC further support the model's strong performance, correctly identifying 93% of positive cases.