LSMA-PQR: Lumbar Spine Multi-view Annotations with Pfirrmann grading, Quantitative measurements and structured Radiological reports

A clinically curated multi‑view lumbar MRI dataset for segmentation, grading, and quantitative analysis. Contains T1‑ and T2‑weighted axial and sagittal scans for 515 patients (4,120 images total: 3,090 axial, 1,030 sagittal) at 384×384 resolution. Each image is annotated for seven anatomical structures across complementary imaging planes: axial view captures intervertebral discs (IVD), posterior elements (PE), and spinal canal (SC) at three clinically relevant levels (L3-L4, L4-L5, L5-S1), while sagittal view depicts vertebral bodies, intervertebral discs, sacrum, and longitudinal spinal canal in mid-sagittal cross-section. Quantitative degeneration assessment includes: - 1,545 radiologist-validated Pfirrmann grades (515 patients × 3 disc levels) representing standardized MRI-based disc degeneration severity (Grade 1-5 scale) - 1,545 intervertebral disc height measurements with pixel-coordinate provenance (mean: L3-L4 = 10.32 ± 1.87 mm; L4-L5 = 10.15 ± 2.14 mm; L5-S1 = 8.98 ± 2.45 mm) - Structured clinical metadata extracted from 515 radiologist reports following systematic error correction (340 transcription/semantic errors corrected across 59% of reports) Features - The dataset enables supervised training for anatomical structure delineation with 100% annotation completeness. Researchers should note that class IDs are view-specific and non-overlapping (axial: 0-2; sagittal: 3-6), requiring conditional class masking in multi-view fusion architectures to prevent cross-view misclassification. - Pfirrmann grades represent ordinal categorical data (1 < 2 < 3 < 4 < 5) where prediction errors should be penalized proportionally to grade distance (predicting Grade 2 when truth is Grade 5 is clinically worse than predicting Grade 4). - The dataset supports structure-symptom analyses by linking pixel-level anatomy to narrative pathology mentions (disc bulge: 53.7%; compression: 78.3%; herniation: 13.9%) and severity categorizations (mild: 34.3%; moderate: 22.8%; severe: 19.1%). - It also supports co-registration of segmentation masks, degeneration grades, disc heights, and clinical descriptors. Standardized Filenames - Axial: {Modality}{PatientID}{DiscLevel}.png (e.g., T1_0042_D4.png) - Sagittal: {Modality}{PatientID}_S{SliceNumber}.png (e.g., T2_0001_S8.png) Dataset Contents - Images Folder : Containing 4120 Images (3090 x Axial and 1030 x Sagittal Images in .png format) - Labels in YOLO format. (4,120 YOLO .txt annotations) - Masks (4,120 semantic segmentation masks) - Overlay Visuals (4,120 quality control overlays (PNG)) - Pfirrmann Grading (.csv file) - IVD Heights (.csv file) - Radiological Notes (.csv file) Our Dataset is free to use with a request to cite our work.本数据集为经临床精心整理的多视角腰椎磁共振成像（Magnetic Resonance Imaging, MRI）数据集，适用于解剖结构分割、病变分级与定量分析任务。 数据集包含515名患者的T1加权（T1-weighted）与T2加权（T2-weighted）轴位及矢状位扫描图像，总计4120张：其中轴位图像3090张，矢状位图像1030张，分辨率均为384×384。每张图像针对互补成像平面中的7种解剖结构完成标注：轴位视图在三个临床相关节段（L3-L4、L4-L5、L5-S1）覆盖椎间盘（intervertebral discs, IVD）、椎后部结构（posterior elements, PE）及椎管（spinal canal, SC）；矢状位视图则在正中矢状切面中显示椎体、椎间盘、骶骨及纵向椎管。 定量退变评估内容包括： - 1545份经放射科医师验证的费氏分级（Pfirrmann grades），对应515名患者×3个椎间盘节段，采用1~5级标准化MRI椎间盘退变严重程度评分体系； - 1545份椎间盘高度测量数据，附带像素坐标溯源信息，各节段均值分别为：L3-L4节段10.32±1.87 mm、L4-L5节段10.15±2.14 mm、L5-S1节段8.98±2.45 mm； - 经系统误差校正后从515份放射科报告中提取的结构化临床元数据：共修正59%报告中存在的340处转录与语义错误。 数据集特性： - 本数据集支持解剖结构轮廓勾画的监督训练，标注完整性达100%。研究者需注意，类别ID与视图相关且无重叠（轴位视图：0~2；矢状位视图：3~6），在多视图融合架构中需使用条件类别掩码以避免跨视图分类错误； - 费氏分级为有序分类数据（1 < 2 < 3 < 4 < 5），预测误差应根据分级差值进行比例化惩罚，例如当真实分级为5时预测为2，其临床危害程度远高于预测为4的情况； - 本数据集支持结构-症状关联分析：可将像素级解剖信息与叙事性病理描述（椎间盘膨出：53.7%；压迫：78.3%；椎间盘疝出：13.9%）及严重程度分类（轻度：34.3%；中度：22.8%；重度：19.1%）进行关联； - 该数据集同时支持分割掩码、退变分级、椎间盘高度及临床描述信息的共配准。 标准化命名规则： - 轴位图像：{Modality}{PatientID}{DiscLevel}.png（示例：T1_0042_D4.png） - 矢状位图像：{Modality}{PatientID}_S{SliceNumber}.png（示例：T2_0001_S8.png） 数据集内容： - 图像文件夹：包含4120张.png格式图像（3090张轴位图像与1030张矢状位图像）； - YOLO格式（YOLO format）标注：共4120份YOLO .txt格式标注文件； - 分割掩码：共4120张语义分割掩码； - 叠加可视化图：共4120张质量控制叠加图像（PNG格式）； - 费氏分级数据：.csv格式文件； - 椎间盘高度数据：.csv格式文件； - 放射科报告：.csv格式文件。 本数据集免费开放使用，敬请引用本研究成果。