Decoding the Rotation Effect: A Retrospective Analysis of Lesion Orientation and Its Impact on Wavelet-Based...

Area of Science:

• Radiomics
• Medical Imaging Analysis
• Computational Pathology

Background:

• Wavelet decomposition (WD) is a common radiomics technique.
• WD is sensitive to input data rotation, potentially altering feature values.
• The impact of lesion orientation on WD features and model performance is not well understood.

Purpose of the Study:

• To investigate the effect of lesion orientation variations on WD and non-WD radiomic features.
• To assess the impact of orientation on radiomic model performance for non-small-cell lung cancer (NSCLC) classification.
• To compare the stability of WD versus non-WD features under rotation.

Main Methods:

• CT radiomics data from 419 NSCLC patients were analyzed.
• Random rotations (5°–80°) were applied to tumors before feature extraction.
• Feature value differences and model accuracy changes were quantified and statistically validated.

Main Results:

• 23.7% of WD features showed significant correlation with rotation, versus 0.5% for non-WD features.
• WD-based models demonstrated significant performance degradation with rotation (Spearman's CC = -0.44, p < 0.001).
• Non-WD models showed no significant performance change with rotation (Spearman's CC = 0.03, p = 0.07).

Conclusions:

• Lesion orientation significantly affects radiomic feature values and model reproducibility.
• WD features are less stable than non-WD features when subjected to orientation variations.
• Radiomic study protocols must account for lesion orientation to ensure reliable and reproducible results.