Using Artificial Intelligence to Diagnose Osteoporotic Vertebral Fractures on Plain Radiographs

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Osteoporotic vertebral fracture (OVF) poses significant risks to the elderly population, with current diagnostic methods facing challenges in accuracy and efficiency.
• High rates of misdiagnosis and underdiagnosis, coupled with substantial radiologist workload, necessitate innovative diagnostic solutions.
• Plain radiography offers a simple, fast, and cost-effective imaging modality for OVF assessment.

Purpose of the Study:

• To develop and validate a deep learning-based system (AI_OVF_SH) for the accurate diagnosis and grading of vertebral fractures using plain radiographs.
• To assess the system's performance in skeletal position detection, segmentation, and fracture identification and grading.
• To evaluate the AI system's diagnostic accuracy, sensitivity, and specificity in both internal and external validation datasets.

Main Methods:

• Development of a deep learning model incorporating a multitasking network for skeletal detection, segmentation, and vertebral fracture assessment using area loss ratio.
• Training and internal validation using 11,397 plain radiographs from six Shanghai community centers.
• External validation with 1276 participants from Shanghai Sixth People's Hospital, with diagnoses confirmed by radiologists using the Genant semiquantitative tool.

Main Results:

• The AI_OVF_SH system demonstrated high accuracy and computational speed in skeletal detection and segmentation.
• Internal validation showed high accuracy (97.41%), sensitivity (84.08%), and specificity (97.25%) for all fractures.
• External validation confirmed robust performance with accuracy (96.85%), sensitivity (83.35%), and specificity (94.70%), with high sensitivity and specificity for moderate and severe fractures.

Conclusions:

• The AI_OVF_SH system is an effective tool for assisting radiologists and clinicians in diagnosing vertebral fractures.
• The deep learning approach shows promise in improving the accuracy and efficiency of OVF diagnosis on plain radiographs.
• The system's validated performance suggests its potential to enhance patient outcomes by enabling timely and accurate OVF detection.