Fine-Grained Rib Fracture Diagnosis with Hyperbolic Embeddings: A Detailed Annotation Framework and Multi-Label Classification Model

TL;DR

The paper tackles the lack of fine-grained rib fracture annotations and the need to align imaging with clinical narratives. It introduces a fine-grained annotation framework and a two-stage detection plus hyperbolic multi-modal model that maps CT patches and radiology text into a shared hyperbolic space. The approach yields improved recall on AirRib and RibFrac datasets and demonstrates that textual descriptors meaningfully boost performance, enabling more accurate RibScore estimation for treatment planning. This framework advances automated rib fracture analysis and holds potential for broader clinical adoption and extension to other anatomies with hierarchical fracture taxonomy.

Abstract

Accurate rib fracture identification and classification are essential for treatment planning. However, existing datasets often lack fine-grained annotations, particularly regarding rib fracture characterization, type, and precise anatomical location on individual ribs. To address this, we introduce a novel rib fracture annotation protocol tailored for fracture classification. Further, we enhance fracture classification by leveraging cross-modal embeddings that bridge radiological images and clinical descriptions. Our approach employs hyperbolic embeddings to capture the hierarchical nature of fracture, mapping visual features and textual descriptions into a shared non-Euclidean manifold. This framework enables more nuanced similarity computations between imaging characteristics and clinical descriptions, accounting for the inherent hierarchical relationships in fracture taxonomy. Experimental results demonstrate that our approach outperforms existing methods across multiple classification tasks, with average recall improvements of 6% on the AirRib dataset and 17.5% on the public RibFrac dataset.

Figures (3)

• Figure 1: CT scans showing different types of rib fractures: (a) Buckle fracture in the anterior Right rib, (b) Non-displaced fracture in the posterior left rib, (c) Displaced oblique fracture in the lateral right rib, and (d) Severely displaced fracture.
• Figure 2: Flowchart illustrating the classification of rib fractures based on various parameters, including location, type (displacement), fracture characterization, multiple fractures, rib side, and rib number.
• Figure 3: Overview of our proposed two-stage framework for fine-grained rib fracture analysis. The input CT scan ($X_i$) is processed by the detection model ($D_\theta$) to locate potential fracture patches ($p_i$). These patches are classified ($F_\theta$) for four distinct characteristics: Location, Type of Fracture, Multiple Fractures, and Fracture Characterization. Radiological descriptions are mapped alongside visual features to enable fine-grained classification.