Objective: To differentiate patients with esophageal cancer or premalignant lesions from the high-risk population for preliminary screening of esophageal cancer using a feature index determined by a computer-aided tongue information acquisition and processing system (DS01-B).
Methods: Totally, 213 patients diagnosed with esophageal cancer or premalignant lesions and 2,840 normal subjects were collected including primarily screened and reexamined, all of them were confirmed with histological examinations. Their tongue color space values and manifestation features were extracted by DS01-B and analyzed. Firstly, the analysis of variance was performed to differentiate normal subjects from patients with esophageal cancer and premalignant lesions. Secondly, the logistic regression was conducted using 10 features and gender, age to get a predictive equation of the possibility of esophageal cancer or premalignant lesions. Lastly, the equation was tested by subjects undergoing primary screening.
Results: Saturation (S) values in the HSV color space showed significant differences between patients with esophageal cancer and normal subjects or those with mild atypical hyperplasia (P<0.05); blue-to-yellow (b) values in the Lab color space showed significant differences between patients with esophageal cancer or premalignant lesions and normal subjects (P<0.05). Logistic regression analysis showed that the computer-aided tongue inspection approach had an accuracy of 72.3% (2008/2776) in identifying patients with esophageal cancer or premalignant lesions for preliminary screening in high-risk population.
Conclusion: Computer-aided tongue inspection, with descriptive and quantitative profile as described in this study, could be applied as a cost- and timeefficient, non-invasive approach for preliminary screening of esophageal cancer in high-risk population.
Keywords: computer-aided tongue inspection; esophageal cancer; logistic regression; preliminary differential diagnosis.