CaviScanNet: AI-Powered Cavity Detection, Segmentation, and Diagnosis with BERT Recommendations
Aruna Vipparla1,a) and Siva Prasanth Mysetla2,b), Nithin Paidi3,c), Sree Lakshmi Sathvika Mallempati 4,d), Esther Rani Karnati5,e)
1Assisstant Professor, Department of Computer Science and Engineering, NRI Institute of Technology, Agiripalli-521212, Vijayawada, Andhra Pradesh, India
2,3,4,5B. Tech Student, Department of Computer Science and Engineering, NRI Institute of Technology, Agiripalli-521212, Vijayawada, Andhra Pradesh, India
a)Corresponding Author: aruna.vipparla5@gmail.com
b)msivaprasanth5599@gmail.com
c)paidinithin2004@gmail.com
d)sree.malempati08@gmail.com
e)esther1323.com@gmail.com
Abstract.This paper introduces a deep learning-based system for dental X-ray analysis aimed at automating cavity detection, severity classification, and providing personalized recommendations. Using Mask R-CNN, the system detects cavities and segments their affected areas, while ResNet-50 classifies the severity of caries into superficial, medium, or deep categories. A fine-tuned BERT-based recommendation system then offers tailored advice based on severity and potential causes such as poor hygiene or diet. The solution reduces manual diagnostic effort, enhances accuracy, and provides actionable insights, which can be deployed via a web interface for remote accessibility and clinical integration, thus advancing dental care and early intervention.The results shows that the detection is highly accurate with 89.2% mAP (Mean Average Precision), the segmentation accuracy by DSC (Dice Similarity Coefficient) was 91.5%, the classification produced 92% validation accuracy among the superficial, medium, deep caries and recommendation had a 90% relevance score matched with dentists advice which is a one-of-a-kind feature.
Keywords: Deep learning, Mask R-CNN, ResNet-50, Image segmentation, Feature Extraction, BERT-based Recommendation System, Mean Average Precision, Dice Similarity Coefficient
