CFPNet-M: A light-weight encoder-decoder based network for multimodal biomedical image real-time segmentation.

Area of Science:

• Medical image analysis
• Deep learning
• Computer vision

Background:

• Deep learning, particularly U-Net, is dominant in medical image segmentation.
• Existing U-Net models are computationally intensive and may struggle with complex segmentations.
• Limitations include high parameter counts, substantial memory usage, and lack of global context.

Purpose of the Study:

• To develop a more efficient and effective deep learning model for medical image segmentation.
• To address the limitations of existing U-Net architectures, such as high computational cost and parameter count.
• To introduce a novel lightweight architecture for improved medical image segmentation.

Main Methods:

• Proposed a novel lightweight architecture: Channel-wise Feature Pyramid Network for Medicine (CFPNet-M).
• Incorporated two modifications: a dilated channel-wise CNN module and a simplified U-shape network.
• Evaluated on five diverse medical imaging datasets (thermography, electron microscopy, endoscopy, dermoscopy, digital retinal images).
• Compared performance against various existing models using Tanimoto similarity for evaluation.

Main Results:

• CFPNet-M achieved segmentation results comparable to existing methods across all five datasets.
• The model requires only 8.8 MB of memory and 0.65 million parameters (approx. 2% of U-Net).
• Demonstrated real-time application suitability with an inference speed of 80 frames per second on a single GPU.

Conclusions:

• CFPNet-M offers a highly efficient and effective solution for medical image segmentation.
• The lightweight design makes it suitable for resource-constrained environments and real-time applications.
• This novel architecture significantly reduces computational and memory demands while maintaining high segmentation accuracy.