Fully Automated, Quality-Controlled Cardiac Analysis From CMR: Validation and Large-Scale Application to Characterize...

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• Deep learning (DL) algorithms show promise for automating cardiac function assessment from cine cardiac magnetic resonance (CMR) imaging.
• Current limitations include variable image quality, diverse disease phenotypes, and DL algorithm training weaknesses, hindering clinical adoption.

Purpose of the Study:

• To develop a fully automated framework for cardiac function analysis from CMR.
• Incorporate comprehensive quality control (QC) algorithms to identify and mitigate erroneous results.

Main Methods:

• A framework integrating pre-analysis DL image QC, DL-based biventricular segmentation, and myocardial feature-tracking (FT).
• Post-analysis QC was implemented to detect erroneous outputs.
• Validation involved comparison with manual analysis (n=100) and QC performance evaluation (n=700), followed by derivation of reference values from UK Biobank data (n=2,029).

Main Results:

• Automated analysis demonstrated high correlation with manual measurements for ventricular volumes (r > 0.95) and strain (r > 0.89).
• Minimal bias was observed for most parameters, with high sensitivity (93-95%) in detecting erroneous outputs.
• Reference values for cardiac function metrics were successfully derived from a large UK Biobank cohort.

Conclusions:

• A DL-based framework enables automated, quality-controlled cardiac function characterization from cine CMR.
• The system operates effectively without requiring direct clinician oversight, paving the way for wider clinical application.