QueenBee-Spine-Detector

YOLOv8x fine-tuned for spine pathology detection on VinDr-SpineXR

Part of the TrustCat sovereign medical AI stack.

Model Description

Detects 8 spine pathologies in X-ray images:

Class	Description
Osteophytes	Bone spurs
Disc space narrowing	Degenerative disc disease
Foraminal stenosis	Nerve root compression
Spondylolisthesis	Vertebral slippage
Vertebral collapse	Compression fracture
Surgical implant	Hardware present
Other lesions	Misc pathology
No finding	Normal

Performance

Metric	Value
mAP@50	44.5%
mAP@50-95	22.4%
Precision	49.0%
Recall	46.8%

Training

• Base Model: YOLOv8x (68M params)
• Dataset: VinDr-SpineXR (3,700 train / 412 val)
• Hardware: 2x RTX 5090
• Epochs: 113 (early stopping, best @ 83)
• Training Time: ~1h 45m

Usage

from ultralytics import YOLO

model = YOLO("Trustcat/queenbee-spine-detector")
results = model("spine_xray.jpg")
results[0].show()

Intended Use

• Clinical decision support (not primary diagnosis)
• Radiology workflow assistance
• Research and development

Limitations

• Trained on single dataset (VinDr-SpineXR)
• Not FDA cleared
• Requires clinical validation before deployment

License

Apache 2.0

---

Built with diamond hands by TrustCat - Sovereign Medical AI