减影碘图联合深度学习重建用于改善颞骨病变骨侵犯的评估
Subtraction iodine maps combined with deep learning reconstruction for improved evaluation of bony invasion in temporal bone lesions.
文献信息
| PMID | 42855533 |
|---|---|
| 原文 | 在 PubMed 查看原文 ↗ |
| 发表日期 | 2026 |
| 作者 | Tong Su |
| 作者单位 | Department of Radiology, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, China. |
| 期刊 | European radiology |
| SCI 分区 | Q1 |
| IF | 5.7 |
| 研究类型 | AI/ML · 临床 |
| 所属专科 | 耳科 |
中文摘要
目的： 评估减影碘图（SIM）联合深度学习重建（DLR）在识别颞骨病变患者骨受累方面的能力，并与采用混合迭代重建（HIR）的常规增强CT进行比较。
材料与方法： 这项前瞻性研究连续招募了77例（平均年龄：46.1±18.9岁；35例男性）接受增强颞骨CT扫描的颞骨病变患者。平扫期和增强期图像分别使用HIR和DLR进行重建。用于评估的彩色SIM融合图像通过从增强图像中减去平扫图像获得SIM，然后将其叠加到骨窗图像上生成。对四组图像（增强-HIR、增强-DLR、SIM-HIR和SIM-DLR）的定性评分和定量参数进行计算和比较。
结果： 定量分析表明，与增强-HIR相比，SIM-DLR显著降低了噪声（更低的SD值）并提高了病变的CNR（均p<0.001）。在定性方面，SIM-DLR在整体图像质量、颞骨病变检测置信度、病变轮廓勾画能力以及颞骨受累可视化方面均显著高于常规增强-HIR（均p<0.001），且读者间一致性高（kappa：0.812-0.904）。增强依赖性分析显示，与低增强病变相比，SIM在高增强病变中显示出明显有益的病变检测、轮廓勾画和骨受累可视化评分（均p<0.05）。
结论： 与常规增强-HIR相比，SIM-DLR通过改善图像质量、解剖勾画和骨受累可视化，显著增强了颞骨CT成像，尤其是对于富血供病变。
要点： 问题 由于病变与骨对比度差，在常规增强CT上准确评估颞骨病变的骨侵犯具有挑战性。发现 与常规CT相比，SIM联合DLR改善了图像质量、病变勾画和骨受累可视化。临床相关性 该技术增强了颞骨病变的术前评估，尤其是富血供病变，有助于手术规划，并可能提高治疗精度和结局。
英文摘要
OBJECTIVES: To assess the capability of subtraction iodine maps (SIM) combined with deep learning reconstruction (DLR) to evaluate the identification of bone involvement in patients with temporal bone lesions, compared with conventional enhanced CT with hybrid iterative reconstruction (HIR).
MATERIALS AND METHODS: This prospective study continuously recruited seventy-seven patients (mean age: 46.1 ± 18.9 years; 35 males) with temporal bone lesions who underwent enhanced temporal bone CT scans. Pre-contrast and contrast-enhanced phase images were reconstructed using HIR and DLR. The colored SIM fusion images for evaluation were generated by subtracting the pre-contrast image from the contrast-enhanced image to obtain the SIM, which was then overlaid onto the bone window image. Qualitative scores and quantitative parameters were computed and compared among four groups of images: Enhanced-HIR, Enhanced-DLR, SIM-HIR, and SIM-DLR.
RESULTS: Quantitative analysis demonstrated that SIM-DLR significantly reduced noise (lower SD values) and increased CNR of the lesion compared to Enhanced-HIR (both p < 0.001). Qualitatively, SIM-DLR exhibited significantly higher scores than conventional Enhanced-HIR in overall image quality, temporal bone lesion detection confidence, lesion contour delineation capability, and temporal bone involvement visualization (all p < 0.001), with high inter-reader agreement (kappa: 0.812-0.904). Enhancement-dependent analysis revealed that SIM showed markedly beneficial scores of lesion detection, contour delineation, and bone involvement visualization in high-enhancement lesions, compared to low-enhancement lesions (all p < 0.05).
CONCLUSION: Compared to conventional Enhanced-HIR, SIM-DLR significantly enhanced temporal bone CT imaging by improving image quality, anatomical delineation, and bone involvement visualization, especially for hypervascular lesions.
KEY POINTS: Question Accurately assessing bony invasion by temporal bone lesions is challenging on conventional enhanced CT due to poor lesion-to-bone contrast. Findings SIM combined with DLR improve image quality, lesion delineation, and visualization of bone involvement compared to conventional CT. Clinical relevance This technique enhances preoperative evaluation of temporal bone lesions, particularly hypervascular pathologies, aiding surgical planning and potentially improving treatment precision and outcomes.