Ren Yacheng, Ma Jingchen, Xiong Junfeng, Lu Lin, Zhao Jun
IEEE Trans Biomed Eng. 2017 Aug;64(8):1924-1934. doi: 10.1109/TBME.2016.2631245. Epub 2016 Nov 22.
Computer-aided detection (CAD) systems for computed tomography colonography (CTC) can automatically detect colorectal polyps. The main problem of currently developed CAD-CTC systems is the numerous false positives (FPs) caused by the existence of complicated colon structures (e.g., haustral fold, residual fecal material, inflation tube, and ileocecal valve). This study proposes a CAD-CTC scheme using shape index, multiscale enhancement filters, and radiomic features to address the FP issue.
Shape index and multiscale enhancement filter calculated in the Gaussian smoothed geodesic distance field are combined to generate the polyp candidates. A total of 440 well-defined radiomic features collected from previous radiomic studies and 200 newly developed radiomic features are used to construct a supervised classification model to reduce the numerous FPs.
The proposed CAD-CTC scheme was evaluated on 152 oral contrast-enhanced CT datasets from 76 patients with 103 polyps ≥5 mm. The detection results were 98.1% and 95.3% by-polyp sensitivity and per-scan sensitivity, respectively, with the same FP rate of 1.3 FPs per dataset for polyps ≥5 mm.
Experimental results indicate that the proposed CAD-CTC scheme can achieve high sensitivity while maintaining a low FP rate.
The proposed CAD-CTC scheme would be a beneficial tool in clinical colon examination.
计算机断层结肠成像（CTC）的计算机辅助检测（CAD）系统能够自动检测结直肠息肉。当前开发的CAD-CTC系统的主要问题是，复杂的结肠结构（如结肠袋褶、残留粪便、充气导管和回盲瓣）的存在导致大量假阳性（FP）。本研究提出一种使用形状指数、多尺度增强滤波器和影像组学特征的CAD-CTC方案，以解决FP问题。
将在高斯平滑测地距离场中计算得到的形状指数和多尺度增强滤波器相结合，生成息肉候选区域。从先前的影像组学研究中收集的总共440个定义明确的影像组学特征和200个新开发的影像组学特征用于构建一个监督分类模型，以减少大量的FP。
所提出的CAD-CTC方案在来自76例患者的152份口服对比增强CT数据集中进行了评估，这些患者有103个≥5毫米的息肉。对于≥5毫米的息肉，按息肉计算的敏感度和每次扫描的敏感度分别为98.1%和95.3%，每个数据集的FP率相同，为1.3个FP。
实验结果表明，所提出的CAD-CTC方案在保持低FP率的同时能够实现高敏感度。
所提出的CAD-CTC方案将成为临床结肠检查中的一个有益工具。
