A task-unified network with transformer and spatial-temporal convolution for left ventricular quantification.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate quantification of cardiac function, particularly left ventricular (LV) function, is crucial for diagnosing and managing cardiovascular diseases.
• Current deep learning methods for LV quantification face challenges due to the heart's dynamic anatomical changes and often lack visual analysis.
• Improving the accuracy of LV quantitative assessment remains a significant research objective in clinical practice.

Purpose of the Study:

• To develop a deep learning framework that simultaneously segments and quantifies left ventricular (LV) function with enhanced accuracy.
• To address the limitations of existing methods by incorporating visual-based analysis and handling the heart's dynamic anatomy.
• To provide a more reliable tool for the clinical assessment of cardiac function.

Main Methods:

• A novel deep learning network unifying segmentation and regression tasks using a transformer and spatial-temporal convolution was proposed.
• The segmentation module employs a U-Net-like 3D Transformer to predict anatomical contours.
• The regression module utilizes spatial-temporal representations and segmentation features for quantification, trained with a joint task loss function.

Main Results:

• The proposed framework achieved competitive cardiac quantification metric results on the MICCAI 2017 Left Ventricle Full Quantification Challenge dataset.
• The method successfully produced visualized segmentation results, aiding in subsequent analysis.
• Experimental results demonstrated the effectiveness of the unified segmentation and regression approach.

Conclusions:

• The developed deep learning framework offers an effective solution for accurate left ventricular (LV) quantification.
• The simultaneous segmentation and regression approach, leveraging transformer and spatial-temporal convolutions, enhances the reliability of cardiac function assessment.
• This method provides visualized outputs beneficial for clinical analysis and diagnosis of cardiovascular conditions.