HUMAN-MACHINE COLLABORATION FOR MEDICAL IMAGE SEGMENTATION.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Image segmentation is crucial in medical image analysis.
• Deep learning models achieve high performance but require extensive annotations.
• Current methods face challenges with limited labeled medical data.

Purpose of the Study:

• To develop a semi-supervised segmentation method for medical images.
• To reduce the need for extensive manual annotation in deep learning models.
• To integrate a human-in-the-loop approach for improved segmentation accuracy.

Main Methods:

• Utilized a conditional Generative Adversarial Network (cGAN) framework.
• Employed the GAN generator to synthesize segmentations on unlabeled data.
• Used the GAN discriminator to identify unreliable image slices needing expert review.

Main Results:

• The proposed method achieved performance comparable to state-of-the-art fully supervised techniques.
• Demonstrated effectiveness in slice-level evaluation on a standard benchmark.
• Significantly reduced the requirement for annotated data compared to traditional methods.

Conclusions:

• Conditional GANs offer a viable solution for semi-supervised medical image segmentation.
• The human-in-the-loop strategy effectively guides annotation efforts.
• This approach enables high-performance segmentation with substantially less labeled data.