Adeoye, John, Koohi-Moghadam, Mohamad, Choi, Siu Wai, Zheng, Li Wu, Lo, Anthony Wing Ip, Tsang, Raymond King Yin, Chow, Velda Ling Yu, Akinshipo, Abdulwarith, Thomson, Peter, and Su, Yu Xiong (2023) Predicting oral cancer risk in patients with oral leukoplakia and oral lichenoid mucositis using machine learning. Journal of Big Data, 10 (1). 39.

Preview

PDF (Published Version) - Published Version Available under License Creative Commons Attribution. Download (1MB) | Preview

 

2

400

Abstract

Oral cancer may arise from oral leukoplakia and oral lichenoid mucositis (oral lichen planus and oral lichenoid lesions) subtypes of oral potentially malignant disorders. As not all patients will develop oral cancer in their lifetime, the availability of malignant transformation predictive platforms would assist in the individualized treatment planning and formulation of optimal follow-up regimens for these patients. Therefore, this study aims to compare and select optimal machine learning (ML)-based models for stratifying the malignant transformation status of patients with oral leukoplakia and oral lichenoid mucositis. One thousand one hundred and eighty-seven patients with oral leukoplakia and oral lichenoid mucositis treated at three tertiary health institutions in Hong Kong, Newcastle UK, and Lagos Nigeria were included in the study. Demographic, clinical, pathological, and treatment-based factors obtained at diagnosis and during follow-up were used to populate and compare forty-six machine learning-based models. These were implemented as a set of twenty-six predictors for centers with substantial data quantity and fifteen predictors for centers with insufficient data. Two best models were selected according to the number of variables. We found that the optimal ML-based risk models with twenty-six and fifteen predictors achieved an accuracy of 97% and 94% respectively following model testing. Upon external validation, both models achieved a sensitivity, specificity, and F1-score of 1, 0.88, and 0.67 on consecutive patients treated after the construction of the models. Furthermore, the 15-predictor ML model for centers with reduced data achieved a higher sensitivity for identifying oral leukoplakia and oral lichenoid mucositis patients that developed malignancies in other treatment settings compared to the binary oral epithelial dysplasia system for risk stratification (0.96 vs 0.82). These findings suggest that machine learning-based models could be useful potentially to stratify patients with oral leukoplakia and oral lichenoid mucositis according to their risk of malignant transformation in different settings.

Actions (Repository Staff Only)

Item Control Page