Evaluating the impact of image enhancement on radiomics feature stability in MRI of nasopharyngeal carcinoma (NPC)

Objective: To evaluate the stability of radiomic features derived from different segmentation methods in head and neck MRI of nasopharyngeal carcinoma (NPC), with a focus on the effect of the Histogram Matching Filter (HMF). Methods: A total of 851 radiomic features, including tumor intensity, shape, and texture, were extracted from 30 manually segmented MRI scans. The same scans were also segmented using semi-automatic techniques and further enhanced using a Histogram Matching Filter (HMF) prior to segmentation. Segmentation was performed using a level tracing algorithm by two experienced radiologists. Intraclass correlation coefficients (ICC) were used to assess feature reproducibility and repeatability. Results: Semi-automatic segmentation with HMF demonstrated the highest reproducibility. For T2-weighted images (T2WI), the ICC was 0.990 ± 0.019 (p < 0.005), and for contrast-enhanced T1-weighted images (CE-T1WI), the ICC was 0.987 ± 0.025 (p < 0.005). Conventional semi-automatic segmentation achieved lower ICCs: 0.905 ± 0.073 for T2WI and 0.922 ± 0.063 for CE-T1WI. Manual segmentation showed the lowest reproducibility with ICCs of 0.787 ± 0.134 for T2WI and 0.801 ± 0.131 for CE-T1WI (p > 0.005). Conclusions: Incorporating HMF into the segmentation workflow significantly improves the reproducibility of radiomic features, especially in T2WI. This enhancement supports more consistent and reliable analyses in radiomic studies. Implication of practice: The use of HMF-enhanced segmentation can reduce variability in radiomic feature extraction, promoting greater consistency in clinical decision-making and radiomic research involving NPC.

Text 124349.pdf - Published Version Restricted to Repository staff only Download (5MB) Official URL or Download Paper: https://www.sciencedirect.com/science/article/pii/...

Actions (login required)

View Item