Dual-feature Fusion Attention Network for Small Object Segmentation.

Area of Science:

• Medical Image Analysis
• Artificial Intelligence in Healthcare
• Radiotherapy Planning

Background:

• Accurate medical image segmentation is crucial for radiotherapy planning and diagnosis.
• Manual segmentation is time-consuming, error-prone, and subject to inter-observer variability.
• Existing automatic methods struggle with small object segmentation due to class imbalance and boundary ambiguity.

Purpose of the Study:

• To develop an advanced deep learning model for accurate medical image segmentation, particularly for small objects.
• To address limitations of current convolutional neural networks in segmenting small medical structures.
• To improve segmentation accuracy by enhancing feature representation and boundary delineation.

Main Methods:

• Proposed a Dual Feature Fusion Attention Network (DFF-Net) incorporating a Dual-Branch Feature Fusion Module (DFFM) and a Reverse Attention Context Module (RACM).
• Employed a multi-scale feature extractor to obtain multi-resolution features.
• DFFM aggregates global and local contextual information for feature complementarity; RACM enhances edge texture to combat blurred boundaries.

Main Results:

• DFF-Net demonstrated superior performance in segmenting small medical objects across NPC, ACDC, and Polyp datasets.
• The proposed method achieved higher accuracy compared to state-of-the-art techniques.
• DFF-Net exhibits fewer parameters, faster inference times, and lower model complexity.

Conclusions:

• The DFF-Net effectively improves medical image segmentation accuracy, especially for challenging small objects.
• The dual-branch fusion and reverse attention mechanisms are key to enhancing feature representation and boundary clarity.
• This network offers a more efficient and accurate solution for clinical diagnosis and radiotherapy planning.