Topological data analysis and machine learning for COVID-19 detection in CT scan lung images

Area of Science:

• Medical Imaging
• Data Science
• Computational Biology

Background:

• Pathogenic laboratory testing for COVID-19 has limitations, including a significant rate of false negatives.
• There is a critical need for accurate, parameter-free diagnostic methods for COVID-19 identification, especially using lung imaging.

Purpose of the Study:

• To develop and validate a novel approach for COVID-19 detection in lung images by integrating topological data analysis (TDA) with machine learning (ML).
• To assess the efficacy of persistent homology features extracted from lung images for accurate COVID-19 classification.

Main Methods:

• Extraction of persistent homology features from lung CT scan images to capture topological properties.
• Utilizing machine learning classifiers, including Random Forest and Support Vector Machine (SVM), with TDA features as input.
• Comparative analysis of model performance with and without topological features.

Main Results:

• The Random Forest Classifier achieved an accuracy rate of 97.5% in classifying COVID-19 positive lung images.
• The Support Vector Machine (SVM) model demonstrated a high performance with an Area Under the Curve (AUC) score exceeding 0.99.
• Topological features significantly enhanced the classification performance of the machine learning models.

Conclusions:

• The integration of topological data analysis with machine learning provides a highly accurate and effective method for COVID-19 detection in lung images.
• Persistent homology features are crucial for improving the precision of ML-based diagnostic tools for respiratory diseases like COVID-19.
• This TDA-ML approach offers a valuable alternative to traditional diagnostic methods, addressing limitations such as false negatives.