Machine learning platform helps predict oral cancer risk
Each individual’s status was defined as positive/negative for oral cancers/OPMDs and histological confirmation of epithelial dysplasia (ED) and squamous cell carcinoma (SCC) was performed for positive status.
The follow-up status of those screened negative is tracked through state-linked electronic health records. Information on demographics, habits, lifestyle, and familial risk factors was collected, and expired carbon monoxide levels (in ppm) were assessed by screen.
A deep learning method for oral cancer diagnosis
Input features (n = 40) and histological diagnoses were used to populate 12 machine learning algorithms with an 80:20 training test split applied to the data at random during development develop.
Recursive feature removal with 10-fold cross-validation is used for object selection while aggregation-minimality-oversampling with modified-nearest neighbors is implemented to correct the loss. class balance.
Internal validation was performed with 20% unused data with comparison of outputs using McNemar’s test used for optimal model selection Performance metrics include recall, specificity, and F1 score.