Multi-Scale 3D Cephalometric Landmark Detection Based on Direct Regression with 3D CNN Architectures

Chanho Song ¹, Yoosoo Jeong ², Hyungkyu Huh ¹

⁰Medical Device Development Center, Daegu-Gyeongbuk Medical Innovation Foundation (K-MEDI Hub), Daegu 41061, Republic of Korea.

Diagnostics (basel, Switzerland) +

|

November 27, 2024

View abstract on PubMed

Summary

This study introduces a novel multi-scale 3D convolutional neural network (CNN) for accurate maxillofacial landmark detection in CT scans. The deep learning approach enhances diagnostic precision in cephalometric analysis.

Area Of Science

Medical Imaging
Computer Vision
Deep Learning

Background

Traditional cephalometric analysis relies on 2D radiographs, but 3D imaging offers greater detail.
Automated landmark detection using deep learning is advancing, yet 3D imaging presents computational challenges.
This research addresses the need for efficient and accurate 3D landmark detection in maxillofacial imaging.

Purpose Of The Study

To develop and evaluate a multi-scale 3D CNN for precise maxillofacial landmark detection.
To improve upon existing 3D CNN architectures for cephalometric analysis.
To provide a reliable automated method for anatomical landmark identification in 3D medical images.

Main Methods

A coarse-to-fine framework using a multi-scale 3D CNN with direct regression.
Utilized a clinical dataset of 150 maxillofacial CT scans with 30 annotated landmarks.
Employed global context identification followed by localized 3D patch refinement.

Main Results

Achieved a mean Root Mean Square Error (RMSE) of 2.238 mm, surpassing conventional 3D CNNs.
Demonstrated consistent and reliable landmark detection without failure cases.
Validated the effectiveness of the multi-scale approach in complex 3D data.

Conclusions

The proposed multi-scale 3D CNN framework offers a reliable solution for automated landmark detection in maxillofacial CT.
This method shows significant potential for improving cephalometric analysis and other clinical applications.
The study highlights the efficacy of deep learning in addressing 3D imaging complexities for anatomical analysis.

Keywords:

3D convolutional neural network (CNN) cephalometric analysis cephalometric landmark detection