Evaluation of Nonlinear Pixel-Wise Squaring for Enhancing Low-Contrast Resolution in CT Imaging.

PURPOSE: In hyperacute stroke, early computed tomography (CT) signs are key indicators for diagnosis. However, the earlier the imaging is performed after the onset, the more subtle the blurring of the corticomedullary and insular ribbon boundaries is, making visual identification extremely difficult. This study investigated whether applying nonlinear pixel-wise squaring to CT images could improve low-contrast resolution. METHODS: Using the low-contrast resolution evaluation phantom Catphan 700 (The Phantom Laboratory, Greenwich, NY), four types of images were generated from the acquired CT data: a standard CT image (normal), an image processed with a subtraction of 5 Hounsfield units (HU) from all pixels followed by squaring (Square-5), a squared image without offset (Square), and an image processed with an addition of 5 HU to all pixels followed by squaring (Square+5). Physical evaluations were performed using a task transfer function (TTF), noise power spectrum (NPS), and contrast-to-noise ratio (CNRLO). Visual assessment was also conducted. RESULTS: The subjective visual evaluation revealed that the effectiveness of the non-linear processing was highly dependent on both the scan mode (volume/helical) and the display window setting (wide/narrow). The proposed Square-5 method demonstrated a significant improvement (p < 0.01) in subjective CNR scores for volume scans, regardless of the window setting. However, it showed no significant improvement for helical scans. Notably, the Square (no offset) method showed the strongest results when combined with the narrow window setting. Under this specific condition, the Square method significantly improved all three subjective CNR scores (p < 0.05) without a statistically significant negative impact on perceived noise. CONCLUSION: The non-linear squaring processing improved low-contrast visibility; however, its effectiveness was highly dependent on both the scan mode and the display window settings. Despite the documented increase in physical noise (NPS), the processing did not cause a corresponding perceived increase in noise. This study demonstrated that under specific conditions, the processing can significantly improve low-contrast visibility, both objectively (CNRLO) and subjectively. These optimal conditions were particularly the Square-5 method for volume scans and the offset-free square method combined with a narrow window.