Parameter-efficient adaptation of foundational models for automated myocardial strain analysis

Clinical Snapshot

Key Findings

• P Value: Strain group discrimination described as statistically significant; specific p-values not provided in abstract

• Effect Size: Mean LV Dice score 0.90±0.02 (SAX, 10 training studies); median AEPE 3.08 mm for motion estimation; SAX-LAX radial strain correlation r=0.64; SAX-LAX longitudinal strain correlation r=0.17

• Primary Outcome: LV segmentation Dice score and myocardial strain discrimination between reference and disease groups across five datasets

• Nnt Or Sensitivity: Diagnostic accuracy metrics (sensitivity, specificity, AUC) not reported; no NNT applicable to this methodological study; Dice coefficient of 0.90 as primary segmentation performance indicator

• Confidence Interval: Not reported for any outcome measure

Clinical Application

The low training data requirement (10 studies for SAX domain adaptation) is a practical advantage for smaller centres or novel scanner deployments. However, clinical deployment would require regulatory approval, integration with existing PACS/CMR workflows, prospective validation in target patient populations, and demonstration of non-inferiority to established strain methods. The pipeline's computational requirements are not described. Cardiac MRI with myocardial strain analysis is performed across major Australian tertiary centres and is relevant to the management of cardiomyopathies, chemotherapy-related cardiotoxicity monitoring, and valvular heart disease assessment. The RACGP and CSANZ do not currently provide specific guidelines on automated AI-based strain analysis tools. TGA regulatory approval as a Software as a Medical Device (SaMD) under the ARTG would be required before clinical deployment in Australia. PBS reimbursement for CMR is limited and context-specific; automated strain tools would need to demonstrate cost-effectiveness within the MBS framework (MBS item 63399 and related cardiac MRI items). Australian CMR practice predominantly uses 1.5T and 3T scanners from major vendors; cross-vendor validation would be essential. The data efficiency advantage of this approach could benefit regional and rural Australian centres with smaller CMR volumes. Adult patients undergoing cardiac MRI for assessment of myocardial function, particularly in settings where automated strain analysis is sought across heterogeneous scanner environments or institutions with limited labelled training data

Abstract

Purpose. Automated myocardial strain analysis requires accurate segmentation and motion tracking, but deep learning methods trained on one dataset often fail when applied to images from different scanners or institutions. We investigate how general-purpose foundational models can be efficiently adapted for cardiac strain estimation, establishing the minimal data requirements for robust cross-domain deployment.Methods. We propose CardiacSAM, a lightweight adaptation of the Segment Anything Model (SAM) fine-tuned with Low-Rank Adaptation, for cardiac structure segmentation, paired with uniGradICON for myocardial motion estimation. We compute global strain in cardiac coordinates for both short-axis (SAX) and long-axis (LAX) views.Results. Validated on four public and one private dataset comprising over 900 subjects, CardiacSAM achieved mean left ventricle (LV) Dice scores of0.90±0.02using only 10 training studies per domain for SAX (50 for LAX). Motion estimation achieved median average endpoint error (AEPE) of 3.08 mm on 15 volunteers without known cardiac disease with expert landmark annotations. Strain measurements discriminated significantly between reference and disease groups across all datasets. Comparing SAX and LAX strain in 360 subjects showed moderate correlation for radial strain (r=0.64) but weak correlation for longitudinal strain (r=0.17), suggesting SAX and LAX capture complementary information about cardiac function.Conclusion. Parameter-efficient adaptation of foundational models enables automated cardiac strain assessment with high data efficiency and robust cross-dataset generalization.

References

1. 1.Bernardo, A., Mato, G., Calandrelli, M., Medus, J., & Curiale, A. (2026). Parameter-efficient adaptation of foundational models for automated myocardial strain analysis. Biomedical Physics & Engineering Express. https://doi.org/10.1088/2057-1976/ae7c09