LSMA-PQR: A Comprehensive Dataset of Lumbar Spine Multi-View Annotations with Pfirrmann Grading, Quantitative...

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence in Medicine

Background:

• Automated lumbar spine MRI analysis is limited by a lack of integrated datasets.
• Existing datasets lack comprehensive pixel-level anatomical data, quantitative degeneration markings, and structured radiological reports.

Purpose of the Study:

• To introduce LSMA-PQR, a novel, clinically validated lumbar spine MRI dataset.
• To provide a unified framework with multi-modal annotations for advancing automated spine analysis.
• To enable research in areas like report-guided supervision and integrated modeling of spine conditions.

Main Methods:

• The LSMA-PQR dataset includes 515 patients with dual-plane, dual-sequence MRI scans (4120 images).
• It features pixel-level segmentation masks for seven anatomical structures with verified spatial consistency.
• The dataset integrates Pfirrmann degeneration grades, disc height measurements, and structured reports derived via natural language processing.

Main Results:

• Clinical validation showed peak degeneration at L5-S1 (71%) and 40% prevalence of severe pathology.
• Baseline experiments demonstrated high performance in multi-plane segmentation (mean Dice = 0.946), Pfirrmann grading (κ = 0.51), and findings extraction (F1 = 0.59).
• The dataset enables previously infeasible research, including report-guided supervision and multi-metric cross-validation.

Conclusions:

• LSMA-PQR is a comprehensive resource for training and evaluating automated lumbar spine analysis systems.
• The dataset facilitates research into the integrated modeling of lumbar spine anatomy, degeneration, and clinical findings.
• Open access to LSMA-PQR under CC BY 4.0 license accelerates translational spine imaging research.