SUSAN: segment unannotated image structure using adversarial network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Accurate medical image segmentation is crucial for diagnosis and treatment planning.
• Unannotated datasets pose a challenge for traditional supervised learning methods.
• Developing automated segmentation techniques for diverse MR image datasets is essential.

Purpose of the Study:

• To develop and evaluate a novel segmentation method using a joint adversarial and segmentation convolutional neural network.
• To achieve accurate semantic segmentation on unannotated MR image datasets.
• To assess the performance of the proposed method against existing segmentation and registration techniques.

Main Methods:

• A segmentation pipeline was developed using a joint adversarial and segmentation network, incorporating CycleGAN for unpaired image-to-image translation.
• A joint segmentation network was integrated for enhanced semantic segmentation capabilities.
• The automated segmentation method, SUSAN, was tested on clinical knee MR image datasets for bone and cartilage segmentation.

Main Results:

• SUSAN demonstrated high segmentation accuracy, comparable to supervised U-Net methods (P > 0.05 for most metrics).
• The method significantly outperformed multiatlas (P < 0.001) and direct registration (P < 0.0001) methods.
• SUSAN showed applicability across knee MR images with varying tissue contrasts.

Conclusions:

• SUSAN enables rapid and accurate tissue segmentation across multiple MR image datasets without requiring sequence-specific annotations.
• The joint adversarial and segmentation network approach shows significant potential for advancing medical image segmentation.
• This method offers a promising solution for segmenting unannotated medical imaging data.