超越整体图像学习:用于高级鼻腔疾病分类的解剖学分区深度学习模型.
Song Li1, Xiang-Hai Hu1, Song Luo1,2
1Department of Otorhinolaryngology, The First Affiliated Hospital with Nanjing Medical University, Nanjing, China.
European radiology
|March 15, 2026
概括
解剖学分区的深度学习显著改善了CT诊断的鼻腔疾病. 这种方法与全图像模型相比,增强了病变特征,显示了临床应用的前景.
科学领域:
- 医疗成像医学成像
- 人工智能的人工智能
- 计算病理学计算病理学
背景情况:
- 整体图像深度学习模型用于CT诊断鼻鼻腔疾病,由于解剖异质性,通常表现不佳.
- 这种局限性阻碍了鼻腔和鼻内各种病理的准确诊断.
研究的目的:
- 调查解剖学分区深度学习是否可以提高CT扫描中鼻腔疾病的诊断准确度.
- 为了比较一个解剖学分区模型与一个全图像模型的性能.
主要方法:
- 一个多中心的回顾性研究包括2947个CT检查.
- 在150个CT扫描上进行了13个解剖区域的手动细分.
- 一个nnU-Net v2模型自动分区,并且在细分的子区域上训练了特定疾病的网络.
主要成果:
- 解剖学分区模型实现了0.739的平均子系数.
- 分区模型的平均AUC为0.801,显著超过全图像模型的AUC0.587.
- 在73个诊断标签中,在42个诊断标签中观察到统计学意义上的AUC改善,平均绝对增加为0.214.
结论:
- 解剖学分区的深度学习显著改善了基于CT的鼻腔疾病诊断.
- 这种方法提供了比全图像方法更可靠的病变特征.
- 这些发现表明AI在鼻腔成像中的常规临床实施具有很强的潜力.
相关概念视频
01:29Anatomy of Respiratory System I: Upper Respiratory Tract
6.2K
The upper respiratory tract plays a vital role in the respiratory system, comprising several structures that facilitate air intake and prepare air for the lungs. It also serves as the first line of defense against pathogens and particles. This tract includes the nose and nasal cavity, the oral cavity, the paranasal sinuses, and the pharynx, each with specific functions and features.
Nose and nasal cavity
The nose and nasal cavity represent the main external openings of the respiratory tract....
Nose and nasal cavity
The nose and nasal cavity represent the main external openings of the respiratory tract....
6.2K
01:24Nose and Nasal Cavity
14.4K
The nose is composed of an observable exterior segment (external nose) and an internal segment within the skull known as the nasal cavity (internal nose). The external nose, visible on the face, consists of a framework of bone and hyaline cartilage enveloped in skin and muscle and lined with a mucous membrane. This structure is supported by the frontal bone, nasal bones, and maxillary bone and is supplemented by a cartilaginous framework comprising the septal nasal cartilage, lateral nasal...
14.4K