O-Net

O-Net integrates convolutional neural networks (CNNs) and Swin Transformer architectures to perform simultaneous medical image segmentation and classification by combining local and global feature representations.

Key Features:

• Hybrid Architecture: Encoder combines CNNs for detailed local feature extraction and Swin Transformers for global contextual representation.
• Deep Fusion Mechanism: Decoder fuses outputs from CNN blocks and Swin Transformer layers to synthesize local and global features for refined segmentation.
• Simultaneous Segmentation and Classification: Shared encoder weights enable joint training of segmentation and classification so segmentation-derived features inform classification.

Scientific Applications:

• Multi-organ CT segmentation (Synapse): Evaluated on the synapse multi-organ CT dataset, demonstrating superior segmentation capability compared to existing state-of-the-art methods.
• Skin lesion segmentation and classification (ISIC 2017): Evaluated on the ISIC 2017 challenge dataset, improving lesion delineation and boosting classification accuracy.

Methodology:

The model is trained on large-scale datasets with joint segmentation and classification objectives; the architecture integrates CNNs and Swin Transformers in encoder and decoder stages, with decoder fusion mechanisms and shared encoder weights.