Restrictive Hierarchical Semantic Segmentation for Stratified Tooth Layer Detection

TL;DR

This paper presents a hierarchical semantic segmentation framework for dental imaging that explicitly encodes anatomical structure, improving accuracy and anatomical coherence in tooth layer detection, especially with limited data.

Contribution

It introduces a novel hierarchy-aware segmentation method that couples level-wise predictions with feature conditioning and probabilistic consistency, outperforming existing indirect hierarchy encoding approaches.

Findings

Hierarchical models improve IoU, Dice, and recall for fine-grained dental structures.

The approach yields more anatomically coherent segmentation masks.

Performance gains are especially notable in low-data regimes.

Abstract

Accurate understanding of anatomical structures is essential for reliably staging certain dental diseases. A way of introducing this within semantic segmentation models is by utilising hierarchy-aware methodologies. However, existing hierarchy-aware segmentation methods largely encode anatomical structure through the loss functions, providing weak and indirect supervision. We introduce a general framework that embeds an explicit anatomical hierarchy into semantic segmentation by coupling a recurrent, level-wise prediction scheme with restrictive output heads and top-down feature conditioning. At each depth of the class tree, the backbone is re-run on the original image concatenated with logits from the previous level. Child class features are conditioned using Feature-wise Linear Modulation of their parent class probabilities, to modulate child feature spaces for fine grained detection. A probabilistic composition rule enforces consistency between parent and descendant classes. Hierarchical loss combines per-level class weighted Dice and cross entropy loss and a consistency term loss, ensuring parent predictions are the sum of their children. We validate our approach on our proposed dataset, TL-pano, containing 194 panoramic radiographs with dense instance and semantic segmentation annotations, of tooth layers and alveolar bone. Utilising UNet and HRNet as donor models across a 5-fold cross validation scheme, the hierarchical variants consistently increase IoU, Dice, and recall, particularly for fine-grained anatomies, and produce more anatomically coherent masks. However, hierarchical variants also demonstrated increased recall over precision, implying increased false positives. The results demonstrate that explicit hierarchical structuring improves both performance and clinical plausibility, especially in low data dental imaging regimes.