Deep‑learning based osteoporosis classification in knee X‑rays using transfer‑learning approach.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Osteoporosis Research

Background:

• Osteoporosis causes fractures, particularly in postmenopausal women and older adults.
• Current osteoporosis diagnostic methods are costly and complex.
• Existing deep learning models for bone radiography have limitations in architecture, dataset adaptation, and feature extraction.

Purpose of the Study:

• To develop an affordable and accurate method for early osteoporosis detection using deep learning.
• To address limitations of current diagnostic systems and deep learning models in medical imaging.

Main Methods:

• Utilized the ResNet-50 model with transfer learning for osteoporosis detection in knee X-rays.
• Trained and fine-tuned the ResNet-50 model on a dataset of 372 X-ray images with verified T-score ratings.
• Compared the performance of ResNet-50 against other models, including VGG-16 and CNNs.

Main Results:

• The fine-tuned ResNet-50 model achieved 90% accuracy, outperforming VGG-16 (88%), non-fine-tuned ResNet-50 (83%), ResNet-18 (79%), and a 3-layer CNN (66%).
• ResNet-50 demonstrated high sensitivity and specificity rates.
• The model proved reliable for detecting osteoporosis through deep transfer learning.

Conclusions:

• Deep transfer learning using ResNet-50 provides a significant enhancement for medical imaging systems.
• The ResNet-50 model offers an effective diagnostic tool for healthcare practitioners for early osteoporosis detection and fracture prevention.
• This approach leads to improved patient outcomes through enhanced diagnostic capabilities.