Unsupervised domain adaptation for Covid-19 classification based on balanced slice Wasserstein distance.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• COVID-19 diagnosis relies on accurate image classification.
• Deep learning models require extensive labeled datasets, which are costly and time-consuming to acquire.
• Existing unsupervised domain adaptation methods struggle with conditional class distributions in medical imaging.

Purpose of the Study:

• To develop a novel unsupervised domain adaptation method for COVID-19 X-ray classification.
• To address the challenge of limited labeled data in medical AI.
• To improve the generalizability of deep learning models across diverse COVID-19 X-ray datasets.

Main Methods:

• Proposed a novel unsupervised domain adaptation technique.
• Utilized balanced Slice Wasserstein distance as the core metric.
• Validated the method using multiple standard domain adaptation and COVID-19 X-ray datasets.

Main Results:

• The proposed method effectively captures discriminative and domain-invariant representations.
• Demonstrated superior data distribution matching compared to existing methods.
• Achieved robust performance across cross-dataset experiments.

Conclusions:

• The novel unsupervised domain adaptation method offers a viable solution for COVID-19 X-ray analysis with limited labeled data.
• Balanced Slice Wasserstein distance is effective for handling conditional class distributions in medical imaging.
• The approach enhances the potential for rapid and accurate COVID-19 diagnosis using diverse X-ray datasets.