<![CDATA[Slice Sensitivity Profile (SSP) and its Full Width at Half Maximum (FWHM) are critical indicators of longitudinal resolution and effective slice thickness in computed tomography (CT), forming a cornerstone of quality assurance (QA) protocols. This study introduces a robust and vendor-neutral framework for automated SSP and FWHM measurement using a deep learning-based approach, designed to overcome the limitations of manual, scanner-specific, and phantom-specific methods. A U-Net convolutional neural network was trained on annotated CT phantom images—including AAPM, Catphan, and ACR models—acquired across Philips, Siemens, and GE CT systems with varied slice thicknesses (1.0 mm and 5.0 mm). The pipeline includes automatic stair-step object segmentation, angular correction via Hough Transform, profile extraction, and real-time FWHM computation. Validation against manual measurements demonstrated strong correlation (r > 0.97) and mean absolute errors below 0.2 mm, with no statistically significant differences across stair-step positions (p > 0.05). The system showed excellent repeatability (CV < 1.5%) and reproducibility (CV < 2.5%), even with phantom repositioning and inter-operator variability. Additionally, the framework maintained consistency across all phantom types and scanner brands, confirming its cross-platform reliability and alignment with IEC 61223-3-5 and AAPM performance standards. These results position the proposed method as a generalizable and scalable QA solution, suitable for clinical integration, automated reporting, and longitudinal CT performance monitoring.]]>
