16 slice ct scanner

The model training and evaluation dataset consists of CT and T2 weighted MRI for 56 patients with prostate cancer treated at a large medical institution in the United States from 2017 to 2021. In this dataset, the IPA was segmented by multiple physicians and reviewed and revised by an expert physician. All CT images were acquired using a 16-slice CT scanner (Royal Philips Electronics, Eindhoven, The Netherlands), and these CT images were acquired with voxel size 1.17 mm × 1.17 mm × 2.5 mm. MRI images were acquired with a voxel size of 0.56 mm × 0.56 mm × 3 mm. The MRI was a rigidly registered CT with a fiducial-based registration most appropriate for prostate segmentation. Fourteen patients were used as the test dataset, and the remaining 42 patients were used for cross-validation. The IPA labels were cleaned to ensure uniformity in the extent of the structure to be segmented. The caudal structure limit was determined as 1 cm beyond the primary target (prostate) planning treatment volume (PTV).
An additional 30 patients treated at the same institution from 2020 to 2021 were recruited for two “inexperienced” physicians (no prior IPA segmentation training) to contour with and without the assistance of the AI model as part of the observer study, since this AI model was developed to assist physicians in contouring the IPA. The details of the observer study are described below.

Patients were immobilized in a supine position on a breast board, with both arms above their head and positioned on an arm support device during the simulation. A knee support was placed under the knees to fix the patients' position and improve their comfort.
Free-breathing 4D-CT scans were acquired using a 16-slice CT scanner (Philips Medical Systems, Inc., Cleveland, OH, USA). Respiratory signals were recorded using the Varian Real-Time Position Management (RPM) System (Varian Medical Systems, Palo Alto, CA, USA) by measuring the displacement of infrared markers placed on the epigastric region of the patient's abdomen. GE Advantage 4D software (GE Healthcare, Waukesha, WI, USA) was used to sort the reconstructed 4D-CT images into ten respiratory phases, labeled as 0 to 90% based on triggered signals. Phase 0% denoted the maximum end inspiration (EI), and phase 50% denoted the maximum end expiration (EE). The 4D-CT images were transferred to the Eclipse Treatment Planning System (TPS) (Eclipse 8.6, Varian Medical Systems, Palo Alto, CA, USA) for structure delineation.