Human Gender Prediction Based on Deep Transfer Learning from Panoramic Radiograph Images

Panoramic Dental Radiography (PDR) image processing is one of the most widely used methods for gender determination in forensic medicine. Deep learning models are widely used in automated analysis of radiological images today due to their high processing speed, accuracy and stability. A few approach using transfer learning is proposed to gender-classify PDR images. In this study, DenseNet121 convolutional neural network (CNN) classifier, which is one of the pre-trained Deep learning architectures, was used. The proposed DenseNet121 network has been expanded and fine-tuned with several additional layers before the final layer to increase its ability to understand more complex patterns from data. At the end of this stage, it has been trained with the dental dataset containing PDR images and has become more experienced. K-fold cross validation method is adopted to increase the accuracy of the proposed DenseNet121 model.In this study the best performance was achieved for the 4,800 test datasets with a classification accuracy of 97.25%. The proposed model, along with Grad-CAM based analysis also revealed that the mandible circumference and teeth are the most significant areas to consider in gender classification.