Mitral Valve Anatomy Analysis Using Multimodal Imaging Data

Mitral valve (MV) disease is one of the most prevalent cardiovascular disorders and requires accurate assessment of annular geometry, leaflet morphology, and subvalvular structures for diagnosis, surgical planning, and transcatheter intervention. In current clinical workflows, multiple imaging modalities are used for complementary purposes. Three-dimensional transesophageal echocardiography (3D TEE) serves as the intraoperative reference for evaluating leaflet motion and valve function, but accurate segmentation of thin, deformable leaflets remains difficult due to acoustic noise and shadowing. Cardiac CT provides clearer visualization of annular landmarks and spatial geometry for pre-procedural planning, yet lacks real-time capability. Surgical videos offer direct visualization of leaflet motion and device interaction but require robust tracking and interpretation. Although these modalities are routinely used together in practice, no unified automated pipeline currently exists to provide consistent mitral valve measurements across the full clinical workflow.

Our team has extensive experience in organizing large-scale medical image analysis benchmarks at MICCAI, including eight prior international challenges across cardiac MR, CT, and ultrasound imaging. These efforts have established a mature multi-center data collaboration and evaluation infrastructure and have supported the development of clinically meaningful segmentation, landmark detection, and measurement algorithms. This foundation enables the construction of a reliable multimodal benchmark for mitral valve analysis.

Despite recent technical advances, existing methods for MV analysis remain fragmented and rarely align with real clinical decision-making. Most current approaches are limited to a single imaging modality or a single subtask, such as annulus segmentation or leaflet detection, and therefore cannot provide the integrated anatomical measurements required for intervention planning. As a result, clinicians still rely heavily on manual measurements, which are time-consuming and subject to inter-operator variability. Furthermore, models trained on one modality often generalize poorly to others due to differences in imaging conditions and acquisition protocols, limiting their practical deployment.

To address these gaps, the proposed benchmark establishes a clinically oriented multimodal framework for automated mitral valve anatomy analysis across CT, 3D TEE, and surgical video data. The three tasks are designed to reflect real clinical usage rather than isolated technical objectives. CT-based analysis provides accurate annular geometry for pre-procedural sizing and device selection; 3D TEE segmentation supports intraoperative assessment of leaflet morphology and valve function; and surgical video analysis enables confirmation of leaflet motion and device–tissue interaction during intervention. Together, these tasks aim to support consistent anatomical understanding across pre-operative planning, intra-operative guidance, and post-operative evaluation.

Importantly, the benchmark encourages unified and cross-task modeling strategies. Anatomical priors derived from CT can support TEE-based leaflet analysis, while motion cues from surgical videos provide temporal validation of structural consistency. Such cross-modal integration promotes the development of coherent and generalizable models that better reflect real clinical workflows.

The final outputs of the challenge are clinically interpretable measurements—including commissural width, anteroposterior diameter, saddle height, and leaflet angles—which directly inform routine decision-making in mitral valve repair, transcatheter edge-to-edge repair (TEER), and annuloplasty. By linking algorithm performance to these clinically meaningful metrics, the proposed benchmark aims to improve measurement reproducibility, reduce operator dependence, and support more consistent intervention planning and outcome assessment. Ultimately, this challenge will provide a standardized and clinically grounded evaluation platform to accelerate the translation of multimodal AI methods into real-world cardiac care.