Benchmarking framework for myocardial tracking and deformation algorithms: an open access database.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Anatomy

Background:

• Accurate cardiac motion analysis is crucial for diagnosing and monitoring cardiovascular diseases.
• Existing cardiac motion analysis algorithms require robust validation against standardized benchmarks.
• A MICCAI workshop challenged the medical imaging community to develop and validate such algorithms.

Purpose of the Study:

• To present a benchmarking framework for validating cardiac motion analysis algorithms.
• To evaluate the performance of four different cardiac motion analysis methodologies.
• To compare the accuracy of algorithms across magnetic resonance (MR) and 3D ultrasound (3DUS) imaging modalities.

Main Methods:

• A database comprising 3D tagged MR (3DTAG), cine steady state free precession MR (SSFP), and 3D ultrasound (3DUS) datasets was created.
• Datasets included a dynamic phantom and 15 healthy volunteers, totaling 1158 image volumes.
• Ground-truth motion was established by manually tracking 12 landmarks across the cardiac cycle, with inter-observer variability assessed (median 0.77-0.84 mm).

Main Results:

• Four institutions (MEVIS, IUCL, UPF, INRIA) submitted motion estimates for evaluation.
• Median tracking errors varied by modality and institution, with 3DTAG showing errors between 0.73-1.52 mm and 3DUS between 3.48-4.78 mm.
• SSFP results showed median errors between 3.09-6.18 mm. Strain analysis revealed good agreement, except for radial strain in lower quality images.

Conclusions:

• The developed benchmarking framework effectively validates cardiac motion analysis algorithms.
• The study provides a comprehensive comparison of different algorithms and imaging modalities.
• Results indicate good overall agreement, highlighting the potential of these methods while identifying areas for improvement, particularly in radial strain estimation.