In Vivo Cervical Precancer Classification Through Multifractal Analysis of Spectral Fluctuations in Intrinsic...

Area of Science:

• Biophysics
• Spectroscopy
• Medical Diagnostics

Background:

• Spectral fluctuations in fluorescence are often disregarded as noise.
• However, these fluctuations contain valuable information about the fluorophore microenvironment.
• This information can potentially be used for medical diagnostics.

Purpose of the Study:

• To develop a novel method for extracting spectral fluctuations from intrinsic fluorescence signals.
• To utilize these extracted fluctuations for classifying normal and precancerous patients.
• To evaluate the efficacy of this method in improving diagnostic accuracy.

Main Methods:

• Employed a discrete wavelet transform (DWT) technique to isolate spectral fluctuations.
• Applied inverse DWT after zeroing approximation and detail coefficients to extract fluctuations.
• Utilized multifractal detrended fluctuation analysis (MF-DFA) to characterize signal complexity.
• Performed Random Forest classification using generalized Hurst and Holder exponents.

Main Results:

• MF-DFA revealed stronger multifractality in precancerous signals.
• The Hurst exponent (H) and Hausdorff dimension (Δα) effectively distinguished between normal and precancerous groups.
• Random Forest classification achieved 96% sensitivity, specificity, and accuracy, with an AUC of 0.98.
• The 'bior2.4' mother wavelet function demonstrated optimal performance.

Conclusions:

• Spectral fluctuations contain distinctive features crucial for differentiating between normal and precancerous states.
• The DWT-based extraction and subsequent analysis provide a robust method for enhanced classification.
• This approach holds promise for non-invasive cancer diagnostics using intrinsic fluorescence.