Texture Analysis On Collagen Second Harmonic Generation Images Of Skin Scars

The quantitative method for noninvasive diagnosis of human skin scar tissues has been a challenging problem in medicine. In this study, scar collagen second harmonic generation (SHG) images were analyzed by image processing. We quantitatively discriminated between normal and abnormal scars and characterized the scar collagen SHG images varied with scar duration.In the quantitative discrimination between normal and abnormal scars, texture analysis combining local difference local binary pattern (LD-LBP) operator and wavelet transform was explored to extract SHG image features that are related to the collagen morphology. Based on the extracted parameters including the homogeneity, directionality and coarseness of collagen, the SHG images were classified into normal or abnormal ones by support vector machine. Compared with other texture analysis methods, the proposed method was demonstrated to achieve higher accuracy and more reliable classification. It could provide assistance for physicians in the diagnostic process.In the characterizations of scar SHG images varied with scar duration, this work developed a phase congruency algorithm combined with the above texture analysis method to quantitatively characterize the collagen morphology in normal scars. The phase and texture parameters from the SHG images quantified the collagen directionality, homogeneity and coarseness with scar duration. By merging the phase and texture parameters, the random forest regression was made use of to predict the scar duration. By comparison with other methods, the proposed method was demonstrated to have more reliability to characterize the collagen morphology of scars. It could also help understand the mechanism of scar progression.