STUDY ON THE USE OF SEGMENTATION ROBUST AUTOMATIC THRESHOLD SELECTION FOR SLICE THICKNESS DETERMINATION IN QUALITY CONTROL PROCESSES WITH COMPUTED TOMOGRAPHY PERFORMANCE PHANTOM

• Main Author: Andrienny Natasya, Clarisa
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/77060
• Institution: Institut Teknologi Bandung

CT scan is one of the complex medical imaging systems that presents several risks. Hence, a quality control (QC) program is essential. In the implementation of QC in CT scans, the American Association of Physicists in Medicine (AAPM) CT performance phantom is required. The aim of this final project is to determine the slice thickness of a CT scan image obtained using the AAPM CT performance phantom. The determination of slice thickness in this study employs the robust automatic threshold selection (RATS) segmentation method. Furthermore, this research also aims to examine the effect of slice thickness on the RATS parameter values, which gives results with high accuracy. The data utilized in this study are derived from Nazliah Azzahra's final project, with acquisition parameters of 120 kVp voltage and 200 mA current. The slice thickness variations applied are 2, 3, and 8 mm. Subsequently, the data is processed using the RATS segmentation method. The RATS parameter values are varied using a trial-and-error approach. The parameter value variations used in this study include lambda factors of 3, 4, and 5, minimum leaf sizes of 10, 20, and 30, and noise thresholds of 10, 20, and 30. Upon segmentation, stair-step images are obtained, each featuring three stair steps. In the subsequent steps, three parallel lines are drawn on each stair step for all segmentation results. Profiles of each line are then obtained. The width of these profiles is measured and averaged to determine the slice thickness from the segmentation results. The obtained width from these profiles is compared to the true value with an allowable error limit.The research findings indicate that different slice thicknesses necessitate different RATS parameter values to achieve low error rates. For 2 mm slice thickness, the most accurate result is obtained with RATS parameter variations of lambda factor 5, min leaf size 20, and noise threshold 30, yielding a 6% error. For 3 mm slice thickness, the most accurate outcome is achieved with RATS parameter variations of lambda factor 4, min leaf size 30, and noise threshold 30, resulting in a 0.5% error. Finally, for 8 mm thickness, the most accurate RATS parameter values are lambda factor 5, min leaf size 10, and noise threshold 10, with a 3% error rate. From this study, it can be concluded that the RATS segmentation method can be employed to determine slice thickness, as 94.4% of RATS segmentation results fall below the maximum allowable error limit.