MTAN: A semi-supervised learning model for kidney tumor segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Medical image segmentation is vital for disease diagnosis and treatment planning.
• Deep learning (DL) models show promise but require extensive labeled data and parameter tuning.
• Creating large labeled medical datasets is labor-intensive and time-consuming.

Purpose of the Study:

• To develop a semi-supervised model for accurate segmentation of kidneys, tumors, and cysts in CT images.
• To leverage both labeled and unlabeled data to overcome limitations of scarce labeled samples.
• To improve the efficiency and accuracy of medical image segmentation.

Main Methods:

• An end-to-end semi-supervised learning model, Mean Teacher Attention N-Net (MTAN), was designed.
• MTAN utilizes an AN-Net architecture functioning as both teacher and student models.
• The teacher model guides the student model using unlabeled data to enhance learning quality and reduce overfitting.

Main Results:

• MTAN achieved high segmentation accuracy on the KiTS19 dataset (Dice scores: 0.975 for kidneys, 0.869 for tumors).
• On the KiTS21 dataset, MTAN demonstrated robust performance (Dice scores: 0.977 for kidneys, 0.886 for masses, 0.861 for tumors, 0.759 for cysts).
• The model effectively utilized unlabeled data, mitigating overfitting and improving segmentation quality.

Conclusions:

• The MTAN model offers a robust solution for medical image segmentation, especially with limited labeled data.
• Its semi-supervised approach effectively utilizes unlabeled data, enhancing segmentation accuracy and reducing overfitting.
• Consistent performance across datasets highlights MTAN's reliability for potential clinical applications.