Learn from one data set to classify all - A multi-target domain adaptation approach for white blood cell...

Area of Science:

• Medical Imaging
• Machine Learning
• Computer Science

Background:

• Traditional machine learning struggles with data from different distributions.
• Domain adaptation is crucial for real-world applications where data varies.
• Differences in data collection and internal dynamics hinder cross-domain transfer.

Purpose of the Study:

• To develop a white blood cell (WBC) classification model robust to domain shifts.
• To enable effective knowledge transfer across diverse datasets using domain adaptation.

Main Methods:

• Utilized a single source domain for training.
• Implemented a multi-target domain adaptation strategy.
• Employed data augmentation, data generation, and fine-tuning techniques.

Main Results:

• The model successfully extracted domain-invariant features.
• Achieved high performance across nine different test datasets.
• Reached a multi-target domain adaptation accuracy of 98.09%.

Conclusions:

• The proposed model effectively overcomes domain differences.
• Demonstrated successful adaptation to various target domains.
• Enables rapid classification of unlabeled samples with minimal labeled data.