Empirical Mode Decomposition and Grassmann Manifold-Based Cervical Cancer Detection

Area of Science:

• Biomedical Engineering
• Spectroscopy
• Machine Learning

Background:

• Cervical cancer is a leading cause of female mortality globally.
• Early and accurate detection is vital for effective treatment and improved patient survival.
• Fluorescence spectroscopy offers high sensitivity for detecting cancer-associated biochemical changes.

Purpose of the Study:

• To develop and validate a reliable method for cervical cancer detection using fluorescence spectral signals.
• To explore the efficacy of empirical mode decomposition (EMD) and Grassmann manifold (GM) learning for spectral signal analysis.
• To enhance diagnostic accuracy and reduce computational costs in cervical cancer screening.

Main Methods:

• Fluorescence spectral signals were collected from 110 subjects with various cervical conditions.
• Empirical Mode Decomposition (EMD) decomposed signals into intrinsic mode functions (IMFs) for feature extraction.
• Grassmann manifold (GM) learning and low-rank representation were used for non-linear subspace analysis and dimensionality reduction.
• Mutual information was applied for feature selection, followed by classification using Random Forest (RF) and other machine learning models.

Main Results:

• The combined EMD and GM approach effectively extracted unique spectral features.
• Feature selection using mutual information significantly reduced computational load.
• The Random Forest (RF) classifier achieved a high five-fold cross-validation accuracy of 99% with a standard deviation of 0.02.
• The proposed method demonstrated superior performance compared to other state-of-the-art classifiers.

Conclusions:

• The integration of EMD and GM learning provides a robust framework for analyzing fluorescence spectral data in cervical cancer detection.
• This approach offers a highly accurate and computationally efficient method for early diagnosis.
• The findings suggest significant potential for fluorescence spectroscopy combined with advanced machine learning techniques in improving cervical cancer screening and management.