MARes-Net: multi-scale attention residual network for jaw cyst image segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Jaw cysts are fluid-filled lesions causing various complications.
• Accurate segmentation of jaw cysts in medical images is challenging due to image complexity.
• Existing deep learning methods struggle with precise jaw cyst delineation.

Purpose of the Study:

• To develop an advanced deep learning architecture for improved jaw cyst segmentation.
• To enhance the accuracy and efficiency of jaw cyst identification in medical imaging.

Main Methods:

• Proposed MARes-Net: a multi-scale attentional residual network.
• Incorporated residual connections to optimize encoder-decoder processes.
• Utilized Scale-Aware Feature Extraction Module (SFEM) and Multi-Scale Compression Excitation Module (MCEM).
• Integrated attention gate modules for refined feature map output.

Main Results:

• MARes-Net achieved high performance metrics: 93.84% precision, 93.70% recall, 86.17% IoU, and 93.21% F1-score.
• Demonstrated superior accuracy in delineating and localizing anatomical structures compared to existing models.
• Validated on an original jaw cyst dataset.

Conclusions:

• MARes-Net offers a robust and effective solution for jaw cyst image segmentation.
• The proposed architecture significantly advances the capabilities of deep learning in medical image analysis.
• This method holds promise for improved clinical diagnosis and patient management.