New Features And Prognosis Of Oral Cancer Based On Digital Histopathological Images

Digital histopathological whole slide images provide a new opportunity for computerized quantitative analysis.More and more studies have found that the stroma morphology and the interaction between immune cells and cancer cells in the tumor microenvironment are highly correlated with the prognosis of patients in a variety of tumors.Human Papilloma Virus(HPV+)oral cancer is a common malignant tumor in the head and neck.Currently,there is no good prognostic index for HPV+oral cancer.In this study,computer image analysis and pattern recognition technology were used to conduct feature modeling,feature extraction and screening of the morphology of oral tumor cells and surrounding tissues from high-resolution digital histopathological images,and then image quantitative analysis techniques,such as recurrence risk prediction model were constructed.This model provides objective and quantitative basis and new ideas for histopathological diagnosis and subsequent prognosis prediction,and can provide theoretical basis for the construction of computer-aided oral cancer pathology prognosis prediction system,which is innovative.The specific contents of this study include:(1)to quantitatively extract the nuclear morphology from the digital histopathological whole-slide image,and to measure the degree of interaction between the tumor microenvironment and the cancerous region by using the nuclear morphology characteristics;(2)to extract the same histological features as the tumor area from the digital histopathological images of hematoxylin-eosin(H&E)stained oral squamous cell carcinoma,and to quantify the heterogeneity of the collagen fiber direction in the tumor microenvironment and the interaction between the collagen fiber direction and the tumor area in the tumor microenvironment;(3)feature screening was carried out on the acquired features,and the selected features were used for classifier training,and the classification model was established to verify the feasibility of features as an independent prognostic indicator;(4)a patient prediction model of oral cancer was constructed,and patient labels were generated by cross-validation of the retention method.These risk prediction labels were used for survival analysis to evaluate the performance of the risk model.In this study,we found that,firstly,patients with oral cancer who had a high interaction score between the tumor microenvironment and the cancerous region had a lower risk of recurrence and a longer survival compared with patients with a low interaction(Hazard Ratio,HR(95%CI)=1.76(0.99?3.13),p=0.0352).Secondly,patients in the tumor microenvironment with disordered collagen fibers and high interaction between the stromal region and the tumor region had lower risk and higher survival than patients with consistent collagen fibers and low interaction between the stromal region and the tumor region(HR(95%CI)=3.64(1.85?7.18),p=0.018).In univariate and multivariate survival analysis,the risk model of oral cancer recurrence constructed by image quantitative analysis can significantly distinguish between relapsed and non-relapsed patients.This finding revealed that the interaction between immune cells and cancer cells in the tumor microenvironment,the orientation of collagen fibers in the tumor microenvironment and the interaction between the stromal region and the tumor region quantitatively measured by computer could be used as independent prognostic indicators to guide the treatment course of patients with oral cancer.