CRTSC: Channel-Wise Recalibration and Texture-Structural Consistency Constraint for Anomaly Detection in Medical...

Area of Science:

• Biomedical image analysis
• Medical imaging
• Artificial intelligence in healthcare

Background:

• Unsupervised anomaly detection is crucial for early disease identification in medical imaging.
• Current deep learning methods overlook anomaly shape and position, limiting diagnostic accuracy.
• Medical chest images from diverse sensors present unique challenges for anomaly detection.

Purpose of the Study:

• To develop an effective framework for unsupervised anomaly detection in medical chest images.
• To address limitations of existing methods by incorporating spatial and shape information of anomalies.
• To enhance the performance and generalizability of anomaly detection across different sensor data.

Main Methods:

• Introduced CRTSC framework integrating Channel-wise Recalibration Module (CRM) and Texture-Structural Consistency Constraint (TSCC).
• CRM enhances feature representation by adjusting channel weights and establishing spatial relationships.
• TSCC improves anomaly shape definiteness by optimizing image similarity loss functions.

Main Results:

• CRTSC demonstrated significant performance improvements on public ZhangLab and CheXpert datasets.
• The framework achieved superior anomaly detection compared to state-of-the-art methods.
• The method proved robust and generalizable for sensor-based medical imaging.

Conclusions:

• The CRTSC framework offers a robust and generalizable solution for unsupervised medical image anomaly detection.
• Integrating CRM and TSCC effectively addresses the limitations of previous deep learning approaches.
• This advancement holds promise for improving early disease detection and intelligent health systems.