A computer-based method of segmenting ground glass nodules in pulmonary CT images: comparison to expert radiologists' interpretations

摘要: Ground glass nodules (GGNs) have proved especially problematic in lung cancer diagnosis, as despite frequently being malignant they extremely slow growth rates. In this work, the GGN segmentation results of a computer-based method were compared with manual performed by two dedicated chest radiologists. CT volumes 8 patients acquired multi-slice CT. 21 pure or mixed GGNs identified and independently segmented readers. The is initialized click point, uses Markov random field (MRF) model for segmentation. While intensity distribution varies different GGNs, used MRF adapted each nodule based on initial estimates. This was run three times using points to evaluate consistency. consistency defined overlap ratio (overlap volume/mean volume). points, mean 0.96±0.02 (95% confidence interval mean), significantly higher than inter-observer between radiologists, indicated 0.73±0.04. computer also intra-observer measurements from same radiologist, an 0.69±0.05 (p-value < 1E-05). concordance expert interpretation demonstrated 0.69±0.05. As shown our data, provided observers, accuracy no worse that one physician’s respect another, allowing more reproducible assessment growth.