纹理分析作为检测骶骨脊索瘤放疗后MRI差异的工具
本研究回顾性分析22例骶骨脊索瘤患者放疗前后MRI,对冠状位T1加权图像提取纹理特征并分类,发现前10个纹理特征的线性判别分析模型可正确分类79.5%的肿瘤,前4个特征鉴别放疗前后的敏感度81.8%、特异度77.2%、AUC 84.5%(P=0.003);使用大语言模型分析所有冠状位T1图像显示放疗后灰度共生矩阵特征相关性显著增加(均值放疗前0.92 vs 放疗后0.95,P=0.016)。结论认为纹理分析是可区分放疗前后脊索瘤的定量工具,可能提供治疗反应的生物标志物。
要点
- 前4个纹理特征鉴别放疗前后骶骨脊索瘤的敏感度81.8%、特异度77.2%、AUC 84.5%(P=0.003)。
- 线性判别分析模型基于前10个纹理特征正确分类79.5%的放疗前后脊索瘤。
- 大语言模型分析显示放疗后灰度共生矩阵特征相关性显著增加(均值0.92 vs 0.95,P=0.016)。
核心结论
- 纹理分析是可区分放疗前后脊索瘤的定量工具,可能提供治疗反应的生物标志物。
研究设计与样本
- 回顾性研究,纳入22例接受术前放疗并获得放疗前后MRI的骶骨脊索瘤患者,比较冠状位T1加权图像上肿瘤最大横截面积感兴趣区提取的纹理特征,并使用大语言模型分析所有冠状位T1图像提取纹理特征。
关键影像表现/征象
- 放疗前后冠状位T1加权图像上肿瘤最大横截面积感兴趣区的纹理特征存在差异。
- 前10个纹理特征的线性判别分析模型正确分类79.5%的放疗前后脊索瘤。
- 前4个纹理特征中3个与像素关系相关,1个与像素灰度变化(小波)相关。
- 大语言模型分析所有冠状位T1图像显示放疗后灰度共生矩阵特征相关性显著增加(均值放疗前0.92 vs 放疗后0.95,P=0.016)。
诊断标准与关键数值
- 前4个纹理特征鉴别放疗前后脊索瘤的敏感度81.8%、特异度77.2%、AUC 84.5%(P=0.003)。
- 前10个纹理特征的线性判别分析模型正确分类79.5%的放疗前后脊索瘤。
临床应用与阅片要点
- 作者认为纹理分析可能提供治疗反应的生物标志物。
局限性
- 原文未报告局限性
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
OBJECTIVE: Sacral chordoma is a rare bone tumor frequently treated with pre-surgical radiation; however, treatment response is difficult to determine based on imaging features of conventional MRI. We hypothesized that MRI texture analysis may be capable of demonstrating the effects of radiation therapy.
MATERIALS AND METHODS: We retrospectively examined pre- and post-radiation therapy MRI of patients with sacral chordomas. Firstly, the regions of interest delimiting the largest cross-sectional area of the tumors on coronal T1-weighted images were compared between pre- and post-radiation therapy MRI. Texture analysis features were extracted, analyzed, and classified. Secondly, all the coronal T1-weighted images were compared using large language models (LLM) to extract texture features.
RESULTS: A total of 22 patients received pre-surgical radiation therapy (RT) and obtained pre- and post-radiation therapy MRI. A linear discriminant analysis model of the top 10 texture analysis features classified correctly 79.5% of chordomas pre- vs. post-treatment. The top 4 textural features had 81.8% sensitivity and 77.2% specificity in differentiating pre- vs. post-radiation chordomas, with a receiver operator area under the curve (AUC) of 84.5% (P = 0.003). Three texture features were related to pixel relationships and one to variations in pixel gray level (wavelet). The second approach using all the coronal-T1 images analyzing gray-level co-occurrence matrix (GLCM) features revealed a significant increase in correlation after RT (mean pre, 0.92; mean post, 0.95; P = 0.016).
CONCLUSION: Texture analysis is a quantitative tool capable of differentiating pre- and post-radiotherapy chordomas and may provide biomarkers of treatment response.
引用
- Godoy IRB, Vicentini JRT, Martinez-Salazar EL, et al. Textural analysis as an instrument to detect differences on sacral chordoma MRI post-radiation therapy. Skeletal Radiol. 2025. PMID: 41217469. doi:10.1007/s00256-025-05070-6
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。