MAS-Net:Multi-modal Assistant Segmentation Network For Lumbar Intervertebral Disc

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Lumbar disc disease diagnosis relies heavily on expert interpretation of medical images.
• Current methods for lumbar intervertebral disc (LID) segmentation are time-consuming and subjective.
• Accurate LID segmentation is crucial for intelligent diagnosis but remains challenging due to anatomical complexity.

Purpose of the Study:

• To develop an automated method for accurate segmentation and localization of lumbar intervertebral discs (LIDs).
• To leverage multi-modal lumbar magnetic resonance images (MRIs) for improved LID segmentation.
• To enhance the efficiency and objectivity of lumbar disc disease diagnosis.

Main Methods:

• Proposed a novel multi-modal assistant segmentation network (MAS-Net).
• MAS-Net incorporates a multi-branch fusion encoder (MBFE), cross-modality correlation evaluation (CMCE), channel fusion transformer (CFT), and a selective Kernel (SK) based decoder.
• Utilized skip connections and global pooling for feature enhancement and selective channel weighting.

Main Results:

• MAS-Net achieved high performance with a Dice coefficient of 93.08% on the IVD3Seg dataset and 93.22% on the DualModalDisc dataset.
• The model outperformed existing state-of-the-art networks in LID segmentation.
• Generated accurate 3D models for precise visualization of LIDs.

Conclusions:

• MAS-Net offers an automated solution for LID segmentation, addressing diagnostic challenges.
• Multi-modal MRI integration enhances information complementation and segmentation accuracy.
• The developed network simplifies and clarifies visual representation, aiding medical professionals in diagnosing lumbar disc diseases.