CAT-Net

CAT-Net implements a cross-slice attention Transformer to improve prostate zonal segmentation in magnetic resonance imaging (MRI) by learning multi-scale information from adjacent slices.

Key Features:

• Cross-Slice Attention Mechanism: Learns and exploits multi-scale image-based information from adjacent MRI slices to capture cross-slice context for segmentation.
• Transformer Module Integration: Embeds the cross-slice attention within a Transformer module to process and integrate multi-scale, cross-slice information and operate with skip-connection-based segmentation architectures.
• Improved Segmentation Accuracy: Enhances segmentation accuracy and consistency across prostate slices—apex, mid-gland, and base—with particular improvement in peripheral zone delineation.

Scientific Applications:

• Prostate Cancer Diagnosis and Treatment Planning: Provides more reliable prostate zonal segmentation in MRI to support diagnostic imaging and treatment planning for prostate cancer, a leading cause of cancer death among men in the United States.

Methodology:

Implements a cross-slice attention mechanism that learns multi-scale image-based information from adjacent MRI slices and embeds this mechanism in a Transformer module for use in deep-learning segmentation architectures employing skip connections.