Improved Panoramic Dental X-ray Segmentation Using MSOA-SE-Res-Dense U-Net++ Architecture

Abstract

Panoramic dental X-rays are the most common imaging source for oral conditions; however, their reading is often delayed by artefacts, and poor contrast. A precise segmentation of dental structures represents the core for clinical diagnosis and treatment planning; yet, current solutions frequently fail to generalize and be robust. To address this issue, introduce a novel MSOA-SE-Res-Dense U-Net++ architecture integration the features of Squeeze-and-Excitation (SE) blocks, residual connections with U-Net ++, dense feature propagation, and nested skip pathways for enhancing the feature and gradient stability. The hyperparameters are tuned through the Mantis Shrimp Optimization Algorithm (MSOA) to faster convergence and better generalization. An exhaustive pre-processing process that comprises resizing, Contrast-Limited Adaptive Histogram Equalization (CLAHE), and intensity normalization is aimed at standardizing the input quality. From the Python software, experimental study using “Collection of Dental X-ray Images for Analysis” dataset demonstrates the effectiveness of approach in terms of accuracy (98.00%), precision (96.50%), recall (95.50%), and F1-score (96.50%), thereby outperforming the existing models The accomplished results signify the promise of the presented framework in giving dependable clinical segmentation for dental diagnostics and treatment planning.