UNSX-HRNet: Modeling anatomical uncertainty for landmark detection in total hip arthroplasty

Jiaxin Wan ¹, Lin Liu ¹, Haoran Wang ²

⁰School of Optoelectronic Science and Engineering, University of Electronic Science and Technology of China, Chengdu, 611731, China.

Computers in Biology and Medicine +

|

October 2, 2025

View abstract on PubMed

Summary

A new deep learning framework, UNSX-HRNet, enhances anatomical landmark detection for total hip arthroplasty (THA) using uncertainty estimation. This AI approach improves accuracy with unstructured radiographic data, aiding surgical planning and evaluation.

Area Of Science

Medical Imaging
Artificial Intelligence
Orthopedic Surgery

Background

Accurate anatomical landmark detection is crucial for total hip arthroplasty (THA) surgical planning and evaluation.
Existing methods struggle with unstructured radiographic data like irregular postures or occluded landmarks, limiting reliability.
Developing robust deep learning frameworks is essential to overcome these limitations.

Purpose Of The Study

To develop an advanced deep learning framework leveraging uncertainty estimation for robust anatomical landmark detection in THA.
To address challenges posed by unstructured radiographic data, such as irregular patient postures and occluded landmarks.
To provide clinicians with uncertainty scores for predicted landmarks, guiding focus on critical results.

Main Methods

Proposed Unstructured X-ray - High-Resolution Net (UNSX-HRNet), integrating high-resolution networks with uncertainty estimation.
Developed a method to predict landmarks without a fixed point count, suppressing low-certainty predictions.
Trained and tested the model on both structured and unstructured datasets, evaluating performance with precision metrics.

Main Results

UNSX-HRNet demonstrated significant improvements, exceeding 60% across multiple metrics on unstructured datasets.
The framework maintained high performance on structured datasets, indicating robustness and adaptability.
Uncertainty estimation effectively handled unstructured data and provided certainty levels for landmark predictions.

Conclusions

UNSX-HRNet provides a reliable, automated solution for THA landmark detection, excelling with unstructured data via uncertainty-aware predictions.
The approach enhances accuracy and offers actionable insights for clinicians, supporting AI-driven surgical planning and monitoring.
This work contributes to developing more dependable AI expert systems for orthopedic surgery.

Keywords:

Deep learning Landmark detection Pelvis radiographs Total hip arthroplasty Uncertainty estimation