Abstract
Squamous cell carcinoma (SCC) and high-grade dysplasia (HGD) are two different pathological entities; however, they sometimes share similarities in histological structure depending on the context. Thus, distinguishing between the two may require careful examination by a pathologist and consideration of clinical findings. Unlike previous studies on cancer diagnosis using two-photon microscopy, quantitative analysis or machine learning (ML) algorithms need to be used to determine the subtle structural changes in images and the structural features that are statistically meaningful in cancer development. In this study, we aimed to quantitatively distinguish between SCC and HGD using two-photon microscopy combined with ML. Tissue samples were categorized into two groups: Group 1, primary SCC vs. metachronous HGD (SCC-HGD) and Group 2, primary HGD vs. metachronous HGD (HGD-HGD). We quantitatively analyzed second harmonic generation (SHG) and two-photon fluorescence (TPF) signals from two-photon microscopy imaging of the extracellular matrix (ECM). Gray-level co-occurrence matrix (GLCM) was used to extract the textural features of the tissue images, and support vector machine (SVM), for classification of the tissue images based on their pathologies. The SHG-based classifiers demonstrated 75%, 84.21%, 95%, and 95.65% for Group 1, Group 2, primary SCC vs. primary HGD, and metachronous HGD (Group 1) vs. metachronous HGD (Group 2), respectively. This integrative approach enabled the characterization of different pathological stages and enhances the understanding of the pathogenic factors involved in the progression of esophageal cancer.