STD-Net: a spatio-temporal decoupling network for multiphasic liver lesion segmentation and characterization.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Hepatocellular carcinoma (HCC) is a major cause of cancer mortality.
• Accurate diagnosis via medical imaging is crucial for HCC treatment.
• Current deep learning models often fail to capture temporal dynamics in multiphasic scans.

Purpose of the Study:

• To develop a novel deep learning network, STD-Net, for improved HCC diagnosis.
• To explicitly separate spatial feature extraction from temporal dynamics modeling in medical imaging.
• To enhance the accuracy and robustness of HCC segmentation and characterization.

Main Methods:

• Introduced STD-Net, a spatio-temporal decoupling network.
• Utilized a shared-weight 3D encoder for anatomical representation.
• Employed a transformer-based temporal module for sequential contrast pattern analysis.

Main Results:

• STD-Net outperformed state-of-the-art baselines in segmentation and characterization.
• Achieved higher Dice scores and lower HD95 for HCC.
• Demonstrated superior classification accuracy and stable performance on small or low-contrast lesions.

Conclusions:

• Spatio-temporal decoupling is a promising paradigm for dynamic medical imaging analysis.
• STD-Net offers a more clinically relevant approach to HCC diagnosis.
• The method shows potential for improving diagnostic accuracy in challenging cases.