Self - Supervised Neuro-Symbolic AI for Histopathological Oral Cancer Detection Using Dinov2

Oral squamous cell carcinoma (OSCC) is a major form of oral cancer for which histopathological examination plays an important role in diagnosis.This paper presents a prototype for binary classification of histopathological images into Normal and OSCC categories using a pretrained DINOv2 vision transformer and a trainable classification head.DINOv2 representations are reused from self-supervised pretraining, while the downstream classifier is trained using labelled target images.To support inspection of predictions, the implementation incorporates Grad-CAM visualizations, heuristic image-derived morphology proxies, and a YAML-configured rule engine.The neural prediction and rule-adjusted probabilities are represented separately in the inference output.This distinction is important because a symbolic adjustment does not establish improved classification unless it is explicitly used in a validated fusion procedure.The manuscript describes the architecture, evaluation protocol, required figures, and result-reporting framework.On a strictly held-out test set of 126 histopathological images, the system achieves an overall accuracy of 86.51%, a macro-F1 score of 82.38%, OSCC sensitivity of 89.47%, and an ROC-AUC of 0.9263, demonstrating effective transfer of selfsupervised representations with explainable neuro-symbolic validation.The system is an experimental research prototype and is not a clinically validated diagnostic device.