Catheter Ablation -> Atrial Fibrillation & Atrial Flutter: -> Mapping & Imaging D-AB07 - Alternate Techniques in Energy Sources (ID 6) Abstract Plus

D-AB07-01 - Ultrasound Based 3d Shell Reconstruction Using Deep Learning Technology (ID 1419)


Background: Intracardiac echocardiography (ICE) can be used to create LA geometry before atrial fibrillation (AF) ablation. One of the limitations of current technology is the time it takes to create the LA shell and the operator skills.
Objective: We examined the capabilities and optimal use of the new CARTOSOUND™ FAM algorithm compared to CT images. This new technology is based on deep learning algorithms to automate the 3D shell reconstruction using SOUNDSTAR™ ICE catheters.
Methods: In 47 patients, CARTOSOUND™ LA 3D shells were generated using standard workflow. The new CARTOSOUND™ FAM algorithm provides automatic 3D shell reconstruction and segmentation of the different structures (LA body, LIPV, LSPV, RIPV, RSPV, LAA and esophagus ). The results of the algorithm were compared to CT scans
Results: Average number of clips per patient were 11. The clips were manually annotated at the T wave at average of 2.4 frames per clip. The CARTOSOUND™ FAM algorithm created both 3D shell as well as 2D contours. The algorithm automatically detected the following structures (number represent average number of frames per structure): LA - 15.3, LIPV - 1.4, LSPV - 1.8, RIPV - 2.8, RSPV - 3.2, LAA - 4.5, ESO - 3.1. The average algorithm calculation time was 65 secs. Overall, there was a fair agreement in location of the LA using the two imaging techniques. CT Registration error Average error - 2.7mm STD - 2.1mm respectively. Average Min error - 0mm. Average Max error - 11.97mm.
Conclusion: The SOUNDSTAR™ ICE catheter CARTOSOUND™ FAM algorithm integrated with deep learning technology appear accurate and able to reconstruct with fam the full LA structures. This technology will have several positive clinical implications.