Development and validation of an automated MTF measurement method for five types of Catphan phantoms using bar...

To develop an automated method for measuring spatial resolution using the bar pattern objects in Catphan 500, 503, 504, 600, and 604 phantoms. Method: After identifying the phantom's center coordinates and anchor, we used template matching to estimate the position of each bar pattern object. Subsequently, a circular mask was used to segment the objects locally, yielding the accurate center point of each object. A region of interest (ROI) was then drawn on each object, taking into account its position and angle relative to the phantom center. By plotting the modulation transfer function (MTF) derived from the variance method against the spatial frequency of each object, an MTF curve was constructed. To test the robustness of the proposed method, we evaluated the automated algorithm across phantom model variations, geometric modifications, and reconstruction filters. The measurement results were compared with the visual assessments of 10 human observers. Results: The proposed method successfully identified targets and placed ROIs with 100% accuracy across all phantom models and extreme geometric variations. Statistical validation revealed that the automated 10% MTF limits possess a strong and highly significant positive correlation with the visual detection thresholds of human observers. This robust agreement was consistently maintained across all evaluated phantom models (r = 0.96, p = 0.01) as well as the varying reconstruction filters (r = 0.97, p = 0.03). Under geometric modifications (rotations and flips), the method exhibited excellent measurement stability with a coefficient of variation (CoV) of 4.05%. Conclusion: The proposed automated method accurately measures the MTF of bar pattern objects across Catphan models 500, 503, 504, 600, and 604 with 100% target identification accuracy. It demonstrates geometric robustness against image misalignment, shows high sensitivity in quantifying minor resolution degradation, and correlates strongly with human visual thresholds (r ≥ 0.96). .