Computer-aided detection of equivocal spinal tuberculosis on X-ray using a YOLOv11-based deep learning model

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Spinal tuberculosis (STB), or Pott's disease, is a significant health issue in resource-limited regions.
• X-rays are the primary diagnostic tool, but early STB often shows subtle changes, delaying diagnosis.
• AI-powered analysis of X-rays can help identify suspicious regions for further evaluation.

Purpose of the Study:

• To develop and evaluate a deep learning model for localizing suspicious regions in X-ray images of patients with suspected spinal tuberculosis.
• To assess the model's performance in identifying subtle or equivocal signs of STB that might be missed on initial X-ray review.

Main Methods:

• A retrospective study of 307 patients with equivocal X-ray findings and CT-confirmed STB.
• Development of a YOLOv11-based object detection model for suspicious-region localization on X-ray images.
• Performance evaluation using metrics like mAP, precision, recall, and patient-level detection rates.

Main Results:

• The YOLOv11 model achieved an mAP@0.5 of 0.7664 and a precision of 0.8358.
• The model demonstrated a patient-level detection rate of 0.8073 with a missed-case rate of 0.1927 at a 0.25 confidence threshold.
• Agreement analysis showed good spatial correspondence between AI-identified regions and CT-confirmed STB locations.

Conclusions:

• A deep learning model can effectively localize suspicious regions for STB on X-ray images, even with subtle findings.
• This AI approach offers a feasible framework for early risk prompting and referral support in primary care settings with limited resources.
• Further validation with diverse datasets and prospective studies are needed for real-world clinical application.