A logistic regression model based on tongue image information for prediction precancerous lesions and early stage esophageal cancer in China

Abstract

Background: China is an esophageal cancer high incidence country, with more than 50% esophageal cancer case of the world. Screening and diagnosis of precancerous lesions and early stage cancer are main measures of decreasing incidence rate and mortality rate. Methods: High-risk population (40-69 years old) in Cixian a high risk area of esophageal cancer was screened, and 3053 cases were included. They were devided into 3 groups: normal group, esophageal neoplasia low-level group and esophageal neoplasia high-level group, according to pathology and electronic gastroscope diagnosis. Diagnostic testing lingual information collection system (DS01-B) was used and tongue image were collected, including tongue color, coating color, fur character, tongue shape, local ecchymosis, et al. Difference of tongue image information was analyzed, related clinical variants were analyzed by multi-factor logistic regression. Results: Incidence of local ecchymosis in tongue image was 1.58% (45/2840) in normal group, 2.98% (4/134) in esophageal neoplasia low-level group and 6.32% (5/79)in esophageal neoplasia high-level group. Significant difference was found in 3 groups (P < 0.05). Logistic regression analysis showed that age, male, red tongue, tongue local ecchymosis, yellow and white coated tongue were risk factors for precancerous lesions and early stage esophageal cancer. Logistic regression model was established and this model had diagnostic specificity (80.35%), sensitivity (63.41%) and total coincidence rate (79.51%) for early esophageal cancer screening. Conclusions: Red tongue, tongue local ecchymosis, yellow and white coated tongue were risk factors for precancerous lesions and early stage esophageal cancer. Multi-factor (including tongue image information) logistic regression model has clinical value of prediction precancerous lesions and early stage esophageal cancer.Legal entity responsible for the study: China-Japan friendshop hospital Funding: National "twelfth five"science and technology support plan Disclosure: All authors have declared no conflicts of interest.

From Web to RheumaLpack: Creating a Linguistic Corpus for Exploitation and Knowledge Discovery in Rheumatology

“…Once done, these models could be deployed in a specialised clinic to help the junior rheumatologist summarise the clinical care plan of a report, or could be used as a chatbot for patients seeking answers to their questions after office hours. For instance, authors in Jia et al [2024], developed OncoGPT, this was done by fine-tuning Llama with oncology-related conversations.…”