Jiang Qinling, Sun Hongbiao, Chen Qi, Huang Yimin, Li Qingchu, Tian Jingyi, Zheng Chao, Mao Xinsheng, Jiang Xin'ang, Cheng Yuxin, Wang Yunmeng, Wang Xiang, Wu Su, Xiao Yi
Department of Radiology, Changzheng Hospital, Naval Medical University, Shanghai, China.
Department of Radiology, Kunshan Third People's Hospital, Kunshan, China.
J Thorac Dis. 2025 Jan 24;17(1):289-298. doi: 10.21037/jtd-24-1311. Epub 2025 Jan 22.
Computed tomography (CT) plays an important role in the diagnosis of lung nodules and early screening of lung cancer. The purpose of this study was to compare the efficacy of 1,024×1,024 matrix and 512×512 matrix in an artificial intelligence-based computer-aided diagnosis (AI-CAD) for evaluating lung nodules based on CT images.
This retrospective analysis included 344 patients from two hospitals between January 2020 and November 2023. CT images presenting lung nodules smaller than 30 mm were reconstructed using the 512×512 and 1,024×1,024 matrix. We evaluated image quality and AI-CAD detection of lung nodules. Image quality was subjectively scored using a 5-point Likert method and objectively assessed using image noise and signal-to-noise ratio (SNR). For lung nodules detection, we recorded the accuracy, precision, and recall of AI-CAD for detecting of different types and sizes of lung nodules.
The 512×512 matrix's overall image subjective evaluation score was 3.63, whereas the 1,024×1,024 matrix's was 4.18, among 344 individuals with 4,319 lung nodules. The detection accuracy, precision, and recall of 512×512 and 1,024×1,024 for AI-CAD in all lung nodules were 91.63% 98.32%, 95.68% 98.32%, and 95.59% 100% respectively. Solid, part-solid, and nonsolid nodule identification accuracy on 512 and 1,024 matrix were 91.30% 98.34%, 94.63% 98.50%, and 94.71% 97.74%, respectively, and of <6 mm, 6-8 mm, and >8 mm nodules were 90.58% 97.87%, 96.64% 99.04% and 93.68% 99.36%, respectively.
The 1,024 matrix performed significantly better than the 512 matrix in terms of overall subjective image quality and lung nodule AI-CAD detection rate.
计算机断层扫描（CT）在肺结节诊断和肺癌早期筛查中发挥着重要作用。本研究的目的是比较1024×1024矩阵和512×512矩阵在基于人工智能的计算机辅助诊断（AI-CAD）中基于CT图像评估肺结节的效果。
这项回顾性分析纳入了2020年1月至2023年11月期间来自两家医院的344例患者。对显示小于30mm肺结节的CT图像分别使用512×512和1024×1024矩阵进行重建。我们评估了图像质量以及AI-CAD对肺结节的检测情况。图像质量采用5分李克特法进行主观评分，并通过图像噪声和信噪比（SNR）进行客观评估。对于肺结节检测，我们记录了AI-CAD检测不同类型和大小肺结节的准确率、精确率和召回率。
在344例有4319个肺结节的个体中，512×512矩阵的整体图像主观评估得分为3.63，而1024×1024矩阵的得分为4.18。AI-CAD对所有肺结节中512×512和1024×1024的检测准确率、精确率和召回率分别为91.63%、98.32%，95.68%、98.32%，以及95.59%、100%。512和1024矩阵对实性、部分实性和非实性结节的识别准确率分别为91.30%、98.34%，94.63%、98.50%，以及94.71%、97.74%，对<6mm、6 - 8mm和>8mm结节的识别准确率分别为90.58%、97.87%，96.64%、99.04%，以及93.68%、99.36%。
在整体主观图像质量和肺结节AI-CAD检测率方面，1024矩阵的表现明显优于512矩阵。
