Topology observing 3D device reconstruction from continuous-sweep limited angle fluoroscopy

Area of Science:

• Medical Imaging
• Interventional Radiology
• Surgical Navigation

Background:

• Minimally invasive procedures rely on real-time image guidance, often using 2D fluoroscopy.
• 2D imaging presents challenges in complex anatomical structures like blood vessels and airways.
• Real-time 3D visualization of devices is crucial for improved navigation and procedural outcomes.

Purpose of the Study:

• To develop and evaluate a noniterative 3D device reconstruction algorithm for continuous-sweep limited angle (CLA) fluoroscopy.
• To incorporate endoluminal topology to prevent navigation through disconnected anatomical branches.
• To enhance real-time 3D device visualization during minimally invasive procedures.

Main Methods:

• A static 3D roadmap (RM) of anatomy was used, converted into a graph representation.
• A deep learning approach segmented the device (catheter/guidewire) from live 2D projections, extracting centerlines.
• Depth localization was performed by intersecting device rays with the RM, considering topology to exclude impossible paths.

Main Results:

• The average 3D distance between reconstructed and reference device centerlines was low (specific values omitted).
• The device tip reconstruction accuracy was high (specific values omitted).
• The correct anatomical pathway was identified in a high percentage of frames (specific values omitted).

Conclusions:

• The noniterative 3D reconstruction approach for CLA fluoroscopy is feasible for curvilinear devices.
• This method can enhance navigation during intravascular and transbronchial interventions.
• Improved device navigation is possible in complex and overlapping anatomical structures.