Abstract :
Background: Oral exfoliative cytology serves as a non-invasive diagnostic tool for detecting cellular abnormalities. However, manual analysis can be time-consuming and prone to subjective interpretation. With advancements in Artificial Intelligence (AI), automated systems offer promising solutions for improving diagnostic accuracy and efficiency.Aim: This study aims to evaluate the effectiveness of AI-based techniques, including machine learning and deep learning models, for classifying normal and abnormal oral exfoliative cells through cytomorphological analysis.Materials and Methods: The study employed two AI approaches. The first involved use of cellular and nuclear dimensions, such as cell and nuclear diameters, which were analyzed using a Decision Tree classifier. The second method utilized a deep learning model based on the AlexNet Convolutional Neural Network (CNN) architecture for image-based classification. Grad-CAM visualizations were used to identify biologically significant regions contributing to the classification.Results: The Decision Tree classifier, based on cytomorphometric features, achieved an accuracy of 100% in distinguishing normal from abnormal cells. The AlexNet-based CNN model achieved a classification accuracy of 93%. Grad-CAM results provided interpretability by highlighting relevant morphological areas in the cytological images.Conclusion: The study demonstrates that integrating cytological morphometry with AI techniques significantly enhances the accuracy and interpretability of oral cell abnormality detection. These findings support the potential of AI-assisted cytology as a reliable tool for early screening of oral pathologies.
Keyword :
Decision tree classifier, Deep learning, Morphometric analysis, Exfoliative cytology, Oral malignant disorders