Uncertainty-Guided Active Learning for Access Route Segmentation and Planning in Transcatheter Aortic Valve...

Area of Science:

• Medical Imaging
• Cardiovascular Interventions
• Artificial Intelligence in Medicine

Background:

• Transcatheter aortic valve implantation (TAVI) requires precise selection of the iliac artery for optimal vascular access.
• Manual assessment of iliac artery anatomy for TAVI is time-consuming and can be prone to error.
• Minimally invasive procedures like TAVI necessitate efficient and accurate pre-procedural planning.

Purpose of the Study:

• To develop and validate an active learning-based segmentation framework for contrast-enhanced Cardiac Magnetic Resonance (CMR) data.
• To enable efficient and accurate quantification of aorto-iliac artery diameters for TAVI access route assessment.
• To reduce the manual annotation burden in pre-procedural planning for TAVI.

Main Methods:

• An active learning framework utilizing probabilistic uncertainty and pseudo-labelling for segmentation of contrast-enhanced CMR data.
• Automated pipeline integrating segmentation results for diameter quantification of the aorto-iliac route.
• Ablation study comparing pre- and post-contrast CMR data performance.

Main Results:

• The active learning framework achieved high segmentation accuracy with a Dice score of 0.912.
• Diameter quantification demonstrated a mean absolute percentage error (MAPE) of 4.92%.
• Post-contrast CMR data yielded superior performance compared to pre-contrast data.

Conclusions:

• The developed pipeline provides accurate segmentation and detailed diameter profiles of the aorto-iliac access route.
• This automated approach significantly aids in the assessment of vascular access for TAVI procedures.
• The framework offers an efficient solution for pre-procedural planning, potentially reducing complications associated with TAVI.