文献速递头颈神经CT

筛窦锥形束CT成像作为生物识别工具:用于深度学习身份识别的数据集构建

Ethmoid sinus CBCT imaging as a biometric instrument: dataset creation for deep learning identification.
PubMed 文献 · Eur J Radiol · 2026-03-14

研究构建了565例(男性312例、女性253例,年龄6-74岁)筛骨区域轴位CBCT标注数据集,由专家放射科医师完成标注,并训练CNN模型进行性别分类概念验证;表现最佳的微调ResNet-50模型F1分数为87%,提示筛窦CBCT成像可能作为身份识别相关任务的生物识别标志物。

要点

  • 筛骨及筛窦区域可在CBCT轴位图像上由专家放射科医师完成标注。
  • 微调ResNet-50模型基于筛骨区域CBCT图像进行性别分类的F1分数为87%。
  • 该数据集为法医放射学和深度学习研究提供了可重复使用的资源。

核心结论

  • 筛窦CBCT成像似乎有望作为身份识别相关任务的生物识别标志物,所构建数据集为法医放射学和深度学习研究提供了可重复资源。

研究设计与样本

  • 共收集565例CBCT扫描(男性312例、女性253例),年龄范围6-74岁;所有程序经伦理批准并符合赫尔辛基宣言;由专家放射科医师在轴位切片上标注筛骨区域;训练基于CNN的模型作为性别分类的概念验证。

诊断标准与关键数值

  • 表现最佳的微调ResNet-50模型F1分数为87%。

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

BACKGROUND: Internal cranial structures such as the sphenoid and ethmoid bones, along with their associated sinuses, provide valuable biometric information for human identification, particularly when conventional modalities such as fingerprints, dental records, or DNA are unavailable. Cone Beam Computed Tomography (CBCT) offers a non-invasive imaging modality to visualize these complex anatomical structures.

PURPOSE: To construct an annotated CBCT dataset of the ethmoid bone and evalute its utility for deep learning-based gender classification.

MATERIALS AND METHODS: A total of 565 CBCT scans (312 males and 253 females) with age range 6-74 years were collected. All proceedures were conducted under ethical approval and helsinki compliance. Expert radiologists annotated the ethmoid region across axial slices. A CNN-based model was trained as a proff-of-concept for gender classification.

RESULTS: The best- performing model (fine-tuned ResNet-50) achieved an F1-score of 87% demonstrating strong discriminative potential for ethmoid-based gender classification.

CONCLUSION: The dataset provides a reproduciable resource for forensic radiology and deep-learning research. Ethmoid CBCT iamging show promise as a biometric marker for identifying-related tasks.

来源:Eur J Radiol,查看原文。版权归原作者所有。

引用

  1. Alsalama A, Alsmirat M, Al-Rawi N, et al. Ethmoid sinus CBCT imaging as a biometric instrument: dataset creation for deep learning identification. Eur J Radiol. 2026. PMID: 41864172. doi:10.1016/j.ejrad.2026.112796

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。