对算法进行外部验证,以检测脊椎水平错误标记和自动轮错误.
Tucker J Netherton1, Didier Duprez2, Tina Patel3
1Department of Radiation Physics, Division of Radiation Oncology, University of Texas MD Anderson Cancer Center, United States.
Physics and imaging in radiation oncology
|March 25, 2025
概括
一种新的后处理方法提高了自动化工具在CT扫描中识别和概述脊椎体的准确性. 这种验证对于在临床环境中部署人工智能工具至关重要.
科学领域:
- 医疗成像医学成像
- 放射学中的人工智能
- 机器学习用于医学诊断
背景情况:
- 脊椎体的自动轮工具对于高效的放射学分析至关重要.
- 外部验证对于确保不同医疗机构的AI工具可靠性至关重要.
- 以前的脊椎体自动轮工具需要进一步提高临床使用的性能.
研究的目的:
- 为了外部验证开发的脊椎体自动轮工具.
- 研究一种后处理方法,以提高工具的性能,达到临床上可接受的水平.
- 评估工具的性能指标,包括识别率,轮可接受性和质量保证准确性.
主要方法:
- 外部验证使用来自两个机构的CT扫描 (40来自A,41来自B).
- 脊椎体的自动定位,计数,轮和质量保证查 (C1-L5).
- 将绩效指标与原始培训数据集进行比较,并开发后处理技术.
主要成果:
- 最初的测试显示识别率为83% (A) 和92% (B),与培训数据集相比,性能降低.
- 处理后的调整显著提高了两个数据集的识别率,平均为4% (p < 0.01).
- 调整后的算法显示,对脊椎体定位的准确性有所提高.
结论:
- 机器学习管道的后处理调整提高了脊椎体定位精度,达到临床上可接受的标准.
- 在部署在各种临床环境之前,对机器学习和深度学习工具的外部验证是必不可少的.
- 这项研究强调了人工智能驱动的医学成像工具严格验证的重要性.
相关概念视频
01:18Errors in Taping
506
Errors in taping arise from multiple factors that can significantly impact measurement accuracy in surveying. Misalignment of the tape, often due to human error, is one primary source. A skilled rear tapeman, using a telescope, can help correct alignment by guiding the head tapeman; however, human limitations still lead to small inaccuracies. These errors may include misplacement of pins or inaccurate tape readings due to common visual confusions, such as mistaking a six for a nine. Such...
506
01:12Adjusting a Traverse
477
In the site survey of a four-sided traverse, internal angles are essential to ensure geometric accuracy. The survey revealed that the sum of the measured internal angles was 359 degrees and 48 minutes, which is 12 minutes less than the expected 360 degrees. This discrepancy signals an error likely arising from measurement inaccuracies during the fieldwork.To rectify this error, the adjustment process involved distributing the 12-minute shortfall equally across the four internal angles. By...
477