Sex estimation with convolutional neural networks using the patella magnetic resonance image slices.

Abstract

Conducting sex estimation based on bones through morphometric methods increases the need for automatic image analyses, as doing so requires experienced staff and is a time-consuming process. In this study, sex estimation was performed with the EfficientNetB3, MobileNetV2, Visual Geometry Group 16 (VGG16), ResNet50, and DenseNet121 architectures on patellar magnetic resonance images via a developed model. Within the scope of the study, 6710 magnetic resonance sagittal patella image slices of 696 patients (293 males and 403 females) were obtained. The performance of artificial intelligence algorithms was examined through deep learning architectures and the developed classification model. Considering the performance evaluation criteria, the best accuracy result of 88.88% was obtained with the ResNet50 model. In addition, the proposed model was among the best-performing models with an accuracy of 85.70%. When all these results were examined, it was concluded that positive sex estimation results could be obtained from patella magnetic resonance image (MRI) slices without the use of the morphometric method.