Spatio-Temporal Multi-Task Learning for Cardiac MRI Left Ventricle Quantification.

Area of Science:

• Cardiovascular Imaging and Analysis
• Artificial Intelligence in Medical Diagnostics
• Deep Learning for Quantitative MRI

Background:

• Accurate quantitative assessment of cardiac left ventricle (LV) morphology is crucial for diagnosing cardiovascular diseases.
• Current manual methods for LV quantification are time-consuming, tedious, and prone to observer variability.
• Standardized and automated quantification methods are needed to improve diagnostic accuracy and efficiency.

Purpose of the Study:

• To develop and validate a spatio-temporal multi-task deep learning approach for comprehensive LV morphology quantification.
• To simultaneously assess LV morphology indices, regional-wall thickness (RWT), and cardiac phase (systole/diastole) from 3D Cine-MR images.
• To overcome limitations of manual contouring by providing an automated, accurate, and robust quantification solution.

Main Methods:

• Utilized a 3D spatio-temporal encoder-decoder network for LV segmentation.
• Implemented a multi-task learning framework to regress 11 LV indices and classify cardiac phases.
• Trained and evaluated the model on cine-MR sequences from 145 subjects, comparing against state-of-the-art methods.

Main Results:

• Achieved high prediction accuracy with mean absolute errors (MAE) of 129 mm², 1.23 mm, and 1.76 mm for LV/Myocardium cavity regions and RWTs, respectively.
• Demonstrated strong performance with Pearson correlation coefficients (PCC) of 96.4%, 87.2%, and 97.5% for LV dimensions, RWTs, and cavity regions.
• Reported a phase classification error rate of 9.0%, highlighting the method's robustness across diverse cardiac morphologies and image qualities.

Conclusions:

• The proposed deep learning method offers a robust and accurate solution for quantitative assessment of cardiac LV morphology and function.
• This automated approach significantly reduces inter- and intra-observer variability associated with manual measurements.
• The spatio-temporal multi-task learning framework shows great potential for advancing cardiac MRI analysis and clinical diagnosis.