Categorization of collagen type I and II blend hydrogel using multipolarization SHG imaging with ResNet regression

Area of Science:

• Biomaterials Science
• Medical Imaging
• Computational Biology

Background:

• Distinguishing collagen types I and II is crucial for understanding tissue properties.
• Previous methods using peptide pitch angle and anisotropic parameters required complex, time-consuming model fitting.
• These existing methods showed inconsistencies when analyzing collagen type I and II blend hydrogels.

Purpose of the Study:

• To develop an automated deep learning approach for categorizing collagen type I and II blend hydrogels.
• To improve the accuracy and efficiency of collagen blend analysis compared to traditional methods.
• To investigate the utility of multipolarization second harmonic generation (SHG) imaging with ResNet for this task.

Main Methods:

• Utilized a ResNet deep learning model trained on multipolarization SHG images of collagen type I and II blends.
• Images were acquired at 10° intervals across five distinct blend percentages (0%, 25%, 50%, 75%, 100% type II).
• Compared the performance of models pretrained on polarization versus nonpolarization SHG images.

Main Results:

• The ResNet model pretrained on polarization SHG images achieved a mean absolute error (MAE) of 0.021 for categorization.
• The model pretrained on nonpolarization images had a significantly higher MAE of 0.083.
• The pretrained models demonstrated generalizability, accurately regressing blend hydrogels at intermediate percentages (20%, 40%, 60%, 80% type II).

Conclusions:

• Multipolarization SHG imaging combined with ResNet offers a robust and automated method for analyzing collagen blend hydrogels.
• Deep learning significantly enhances the accuracy and efficiency of collagen type discrimination.
• This approach holds potential for automated extraction of valuable information from complex collagen matrices.