Early Stage Detection of Knee Osteoporosis through X-Ray Imaging with Fine-Tuned VGG19

Abstract

Knee osteoporosis (KO) is a painful condition that is often associated with significant mobility issues and is a typical complication that undermines the strength of bones and increases the chances of fractures. Treatment requires early diagnosis; however, normal methods involve a physical examination and bone mineral density tests may overlook early bone defects. It is a customized VGG19 deep learning model that predicts knee osteoporosis with X-ray images. The VGG19 model that has a high performance in the image classification was altered into binary (healthy versus osteoporotic) classification with the help of the transfer learning. The model with both the Healthy and Osteoporosis category was trained and tested on a dataset of knee X-rays consisting of 1,258 training images, 315 validation images, and 472 testing images. The model performed excellent with a high level of recall, F1-scores, and accuracy of 97.5, which is extremely good with a high level of precision of 0.97. The general training and good generalization were observed in the generalization and confusion matrix with only 12 false negative and no false positives with the confusion matrix reveling the true validation loss although sporadic variations were observed in the training and validation loss. The document confirms the capability of deep learning models particularly VGG19 to automatize the detection of knee osteoporosis and, therefore, improve diagnostics accuracy. This plan can significantly enhance patient outcome since it provides physicians with a powerful tool of early diagnosis.

Ähnliche Arbeiten