Design of an Efficient Deep Learning-based Model for Diagnosis of Osteoporosis using Dental Periapical Radiograph Images

Osteoporosis is a chronic bone disease that tends to make the bone fragile enough to break it, hence its early diagnosis would prove to be helpful for both the patients and the medical practitioners to personalize the treatment. Numerous works have already been done involving deep learning algorithms and pre-trained models to predict osteoporosis via the X-ray datasets of knee etc. while very few research works have explored the diagnosis of osteoporosis using the Dental Periapical Radiograph dataset which is the focus of this work. In this paper, a customized deep learning model with various layers and parameters has been implemented on the Dental Radiograph dataset having three classes— Osteoporosis, Osteopenia, and Normal. Apart from the custom model, transfer learning has also been implemented on the pre-trained models like DenseNet121, DenseNet201, VGG19, and MobileNetV2 for the overall comparison and analysis. The highest test accuracy for the custom models is 95.17% and the highest AUC is 99.59% whereas the highest test accuracy among all the fine-tuned pre-trained models mentioned above is 94.57%, and the highest AUC is 97.68%. It is evident that the custom model’s accuracy and AUC have surpassed the fine-tuned model’s performance, making it seem suitable for deployment in real-world settings and developing other applications.