A Novel Method for Describing Texture of Scar Collagen Using Second Harmonic Generation Images

Abstract

Texture features related to scar collagen second harmonic generation (SHG) images are useful for studying scars; however, current computational analysis methods require extensive computing resources. We designed a local orientation ternary pattern (LOTP) method in the SHG images for the purpose of extracting the characterization. SHG images were generated from human scar tissue samples, with scar age ranging from 2 to 40 years. Depending on the complete texture information of LOTP images, we extracted the Tamura features including coarseness, contrast, directionality, regularity, line-likeness, and roughness. Tamura texture features could be measured for all input patterns to set up a regression model about the age of scars and that give well-distributed results. Generalized boosted regression trees were calculated with the computed data, and R2 and root-mean-square error (RMSE) statistical analysis were used to determine accuracy. Use of the LOTP operator allowed for the maximum extraction and relative importance of Tamura feature data, with roughness being the most important feature and line-likeness being the least important feature. Using the LOTP operator resulted in the highest accuracy assessment of scar characteristics compared to other methods, such as improving local ternary pattern, binary gradient contours, and grey level co-occurrence. Our proposed LOTP method requires less computation time than the extension of LTP and describes SHG images with higher accuracy compared to existing algorithms.

Keywords

• Generalized boosted regression trees (GBRT)
• local orientation ternary pattern (LOTP)
• second harmonic generation (SHG) image
• Tamura texture features.