AI-Based Analysis of Port Catheter Tip Position

Determination of Port Catheter Tip Position Using Artificial Intelligence-Based Automated Image Analysis

This prospective, single-center observational study will evaluate the performance of an artificial intelligence (AI)-based automated image-analysis method that will be developed, internally tested, and locked before prospective enrollment. The method will estimate the position of totally implantable venous access port catheter tips in adults undergoing right internal jugular vein port implantation as part of routine care. The automated estimate will be compared with a prespecified manual measurement method. Reference measurements of the catheter tip and the cavoatrial junction will be obtained from clinically indicated chest computed tomography images when both structures are visible. No additional chest radiography or computed tomography will be performed for research purposes, and the automated output will not guide catheter placement or alter clinical decisions. The primary performance measure will be the mean absolute error of the automated estimate in the prospective cohort.

Before prospective enrollment, a two-dimensional nnU-Net-based multitask model will be developed and internally tested using retrospective chest radiographs from the same center. The model parameters, ensemble, preprocessing steps, and technical output rules will be locked before prospective enrollment. The retrospective development and internal test datasets are not part of the enrollment reported for this record.

The prospective component will include one observational cohort of adults aged 18 years or older undergoing right internal jugular vein port implantation during routine clinical care. Implantation, chest radiography, computed tomography, and all clinical management decisions will remain under the responsibility of the treating clinical team and will not be assigned by the research protocol. The automated output will not be made available for procedural guidance or patient management.

For each evaluable participant, the locked automated method will generate an estimate based on chest radiographic image analysis and a patient-specific Delta adjustment. A prespecified manual A+B measurement method will be applied as a comparator in the same participants. Reference measurements will be derived from the first suitable post-implantation noncontrast chest computed tomography examination obtained for a routine clinical indication, provided that both the catheter tip and the cavoatrial junction are visible and no catheter manipulation occurred beforehand. No additional radiographic or computed tomography examination will be requested for the study.

The primary objective is to quantify the error of the automated estimate using mean absolute error in the prospective cohort. Prespecified secondary analyses will evaluate additional error, agreement, pragmatic accuracy, technical assessability, and repeatability measures. The ±10 mm criterion will be treated as a pragmatic accuracy threshold rather than a clinical safety boundary. The study is a same-center prospective performance evaluation; it is not an external validation and is not designed to assess clinical benefit or to establish that the system is ready for clinical use.

Inclusion Criteria:

• Age 18 years or older.
• Planned placement of a totally implantable venous access port through the right internal jugular vein as part of routine clinical care.
• Availability of a suitable pre-implantation chest radiograph obtained as part of routine clinical care.
• Provision of written informed consent.

Exclusion Criteria:

• Age younger than 18 years.
• Cancellation of the routine clinical port implantation plan.
• Planned or performed port placement through a venous access route other than the right internal jugular vein.
• No suitable pre-implantation chest radiograph available for study assessment.
• Advanced thoracic deformity or mediastinal distortion precluding the required study measurements.
• Inability or unwillingness to provide written informed consent.