This multicenter, retrospective study develops and validates artificial intelligence (AI)-based semantic segmentation algorithms for intraprocedural transesophageal echocardiography (TEE) during Transcatheter Mitral Edge-to-Edge Repair (TEER). Using pooled imaging data from multiple high-volume structural heart centers, the study aims to automate recognition of mitral leaflets and MitraClip components, measure leaflet insertion length in real time, and display clip position and orientation. Algorithm performance will be benchmarked against expert manual annotations.
Transcatheter Mitral Edge-to-Edge Repair (TEER) with the MitraClip device is an established minimally invasive treatment for patients with severe mitral regurgitation who are at high surgical risk. The success of TEER relies heavily on real-time transesophageal echocardiography (TEE) to guide precise clip positioning and leaflet capture. However, intraoperative image interpretation remains highly dependent on operator experience, and variability in image quality, patient anatomy, and the dynamic nature of cardiac structures continue to challenge procedural standardization across centers. This multicenter, retrospective imaging study evaluates whether artificial intelligence (AI)-based semantic segmentation can automate the recognition of mitral valve anatomy and MitraClip device components on intraprocedural TEE images. Previously acquired TEE imaging from adult patients who underwent TEER at multiple participating high-volume structural heart centers will be pooled and analyzed. All data are derived from routine clinical care, and only patients with appropriate consent for research use of their clinical and imaging data are included. The study has three objectives: (1) to develop deep learning models that automatically segment the anterior and posterior mitral leaflets and the MitraClip grippers and arms; (2) to automate real-time measurement of leaflet insertion length during the grasping process; and (3) to integrate three-dimensional imaging with intelligent tracking to display clip position and orientation. By drawing on a multicenter dataset, the study aims to improve the generalizability and robustness of the resulting models across diverse imaging environments, operator practices, and patient anatomies. Algorithm performance will be benchmarked against expert manual annotations using established image segmentation metrics.
Study Type
OBSERVATIONAL
Enrollment
1,500
Fuwai Hospital, Chinese Academy of Medical Sciences
Beijing, Beijing Municipality, China
IRCCS Policlinico San Donato
Milan, Italy
San Raffaele Hospital
Milan, Italy
Accuracy of AI-based semantic segmentation of mitral valve leaflets and MitraClip device components
The accuracy of the deep learning model in segmenting the anterior and posterior mitral leaflets, MitraClip grippers, and clip arms on intraprocedural transesophageal echocardiography (TEE) images. Performance is benchmarked against manual annotations provided by experienced echocardiographers and quantified using the Dice similarity coefficient, sensitivity, and specificity. Target performance: ≥ 90%.
Time frame: Intraprocedural (TEE images acquired during the TEER procedure)
Accuracy of automated real-time recognition of mitral leaflet insertion length
The accuracy of the automated measurement system in recognizing the insertion length of the anterior and posterior mitral leaflets in two-dimensional TEE planes during the leaflet grasping process. Algorithm output is compared with manual measurements performed by experienced echocardiographers. Target performance: ≥ 95%.
Time frame: Intraprocedural (TEE images acquired during the TEER procedure)
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.