Cross-Dimensional Analysis of U-Net Variants for Dental Caries Detection across Multi-Modal Radiographs

View at publisher Summarise with VIRAT
Authors: Tilottama Pankaj Dhake, Namrata Ansari
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This research assesses the performance of several U-Net variants (2D, 2.5D, 3D, 3.5D, Half U-Net, V-Net, Z-Net) for recognizing dental caries in panoramic and CBCT X-rays. The analysis focuses on segmentation accuracy, F1scores, and dataset computational complexity, consisting of 400 annotated images (300 panoramic X-rays, 100 CBCT) sourced from the Narkhede’s Dental Clinic and augmented to 1200 samples. The 2D U-Net performs well on panoramic X-rays, but struggles with CBCT data. 2.5D U-Net improves segmentation accuracy and artifact reduction (F1-score: 0.85) and 3D U-Net achieves maximum accuracy (90.3%) at the highest computational cost. The 3.5D U-Net achieves F1-scores of 0.89 and balances accuracy with computational efficiency. The V-Net, Z-Net, and Half U-Net analyses yield interesting findings. Z-Net, like V-Net, is more powerful than 3D U-Net but not as good as V-Net. Half U-Net offers a more powerful, lowered accuracy option. This study enhances the U-Net framework for segmentation and disease detection in dental image.

Fields of Study & Subjects

Dental Radiography and Imaging Medical Image Segmentation Techniques Medical Imaging and Analysis

Publication details

Journal / source
CRC Press eBooks
Pages
198–204
Publication date
25 March 2026
Language
EN
Document type
Book chapter