PlutoNet: An efficient polyp segmentation network with modified partial decoder and decoder consistency training.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Deep learning models aim to reduce missed polyps and improve segmentation accuracy during interventions.
• Current models struggle with generalization, feature redundancy, and computational demands for real-time applications.

Purpose of the Study:

• To introduce PlutoNet, an efficient deep learning model for polyp segmentation.
• To address the challenges of generalization, feature representation, and computational intensity in polyp detection.

Main Methods:

• PlutoNet utilizes a shared encoder with a novel modified partial decoder and an auxiliary decoder.
• Decoder consistency training enforces feature consistency across different scales and semantic levels.
• The model requires significantly fewer parameters (9 FLOPs, 2,626,537 parameters) compared to existing methods.

Main Results:

• PlutoNet demonstrates superior performance compared to state-of-the-art models, especially on unseen datasets.
• The model achieves high accuracy in polyp segmentation with reduced computational and memory requirements.
• Ablation studies confirm the effectiveness of the proposed decoder consistency training approach.

Conclusions:

• PlutoNet offers an efficient and effective solution for polyp segmentation in medical imaging.
• The novel training strategy enhances feature representation and model generalization.
• This approach holds promise for real-time polyp detection and intervention guidance.