Evaluation of Detection of Saliva and Other Contaminants on Impressions: A Machine Learning Approach for Predictive Accuracy and Automatic Detection


  • Dr. Rahul Koppaka, Dr. Smrithi.N.S, Dr. Ojas A Gajbhiye, Dr Mohika Damle, Nabeel Ahmed*


Machine Learning; Algorithms; Logistic regression; impressions; Silicone; Bayes algorithm; Contamination.


Background And Aim

The application of two distinct machine-learning algorithms for the automated detection of contaminants and fluids in impressions was investigated in this study.

Materials and Methods

Orange, a machine learning tool, was utilized with an algorithm based on squeeze net embedding models. Logistic regression and naive Bayes algorithms were used to train and evaluate prediction and detection algorithms on 27 intraoral pictures of various impressions. The precision was a confusion matrix used to examine the situation.


Three machine learning methods using CNN embedding correctly discovered and predicted the accuracy of impressions. All models demonstrated good accuracy in identifying contaminated impressions. The AUC for Nave Bayes was 0.982, and the AUC for Logistic Regression was 1.000.


With great accuracy, machine learning algorithms can recognize and classify impressions as positive or unfavorable based on contamination. This will help in removing unwanted artifacts while recording intra oral scans or scanning models. More research with larger sample sizes and various imaging modalities is required to confirm these findings