Image analysis framework for infection monitoring

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Accurate monitoring of infection progression in medical images is crucial for effective patient management.
• Existing methods may struggle with image distortions caused by infection manifestations.
• Time-series analysis of medical images offers potential for quantitative assessment of disease dynamics.

Purpose of the Study:

• To develop a novel framework for automatic extraction of infection progress from time-series medical images.
• To apply and validate the framework for pneumonia monitoring.
• To provide a quantitative, time-series output of infection extent.

Main Methods:

• A modified active shape model algorithm, constrained by binary masks, is used for lung detection and delineation.
• Supervised classification of image region samples, represented by fused dissimilarity features, assesses infection extent.
• An entropy-based weighted voting scheme is employed for feature fusion, ensuring nonparametric operation and outlier robustness.

Main Results:

• The framework successfully extracts time-series data quantifying infection progression.
• The proposed method demonstrates improved performance compared to state-of-the-art techniques.
• High overall accuracy of 90.0 ± 2.1% (Area Under the ROC Curve) was achieved for pneumonia monitoring.

Conclusions:

• The developed framework provides an effective and accurate method for monitoring infection progress in medical images.
• The system's robustness, generality, and adaptivity suggest potential applications beyond pneumonia to other medical imaging domains.
• This approach facilitates quantitative, automated assessment of disease dynamics over time, aiding clinical decision-making.