Toward cardiac tissue characterization using machine learning and light-scattering spectroscopy.

Area of Science:

• Biomedical Engineering
• Medical Physics
• Cardiology

Background:

• Non-destructive characterization of cardiac tissue is crucial for surgical planning and evaluation.
• Current methods like OCT and confocal microscopy face barriers such as cost, time delays, and limited scope.
• A rapid, low-cost, non-destructive technique is needed to improve cardiac surgical procedures.

Purpose of the Study:

• To evaluate a novel light-scattering spectroscopy (LSS) system integrated with neural networks for predicting nuclear densities (ND) in cardiac tissues.
• To determine the feasibility of using LSS for non-destructive cardiac tissue characterization.

Main Methods:

• Developed an LSS system with a fiber-optic probe for cardiac tissue measurements in an ovine model.
• Quantified cardiac tissue ND using fluorescent labeling, confocal microscopy, and image processing.
• Analyzed LSS spectra with spectral clustering and convolutional neural networks (CNNs) to assess ND characterization.

Main Results:

• Spectral clustering identified distinct spectral groups correlating with ND ranges.
• CNNs accurately classified spectra into low, medium, and high ND groups with 95.00 ± 11.77% accuracy.
• Classification accuracy demonstrated sensitivity to wavelength range and spectral subsampling.

Conclusions:

• Light-scattering spectroscopy combined with machine learning can effectively assess nuclear density in cardiac tissues.
• This approach shows promise for diagnosing cardiac diseases linked to altered ND, such as hypertrophy and fibrosis.
• The developed LSS system offers a rapid, non-destructive method for cardiac tissue analysis.