Quantification of Protein Secondary Structures from Discrete Frequency Infrared Images Using Machine Learning.

Area of Science:

• Biomedical optics
• Spectroscopy
• Computational imaging

Background:

• Discrete frequency infrared (IR) imaging offers chemical contrast for biomedical applications.
• Analyzing spectral data for protein secondary structure in tissues, crucial for neurodegenerative disease characterization, is computationally intensive.
• Conventional methods like band fitting hyperspectral data are slow and require extensive data acquisition.

Purpose of the Study:

• To develop a computationally efficient method for analyzing discrete frequency IR imaging data.
• To reduce data acquisition time and computational overhead in spectral analysis.
• To enable accurate retrieval of band-fitting results from sparsely sampled spectral data.

Main Methods:

• A two-step regressive neural network model was developed.
• The model interpolates spectral information from a limited number of wavenumbers (seven).
• Upscaled spectra were used to predict component areas under the curve (AUCs).

Main Results:

• Data acquisition time was reduced nearly six-fold.
• The model achieved over 3000x speedup compared to traditional spectral fitting.
• High fidelity in predicting component AUCs was maintained.

Conclusions:

• A neural network approach effectively mitigates computational challenges in discrete frequency IR imaging.
• This method drastically reduces data acquisition and analysis time.
• The approach enhances the potential of discrete frequency imaging for disease characterization by enabling faster protein structure analysis.