[論文レビュー] RegFreeNet: A Registration-Free Network for CBCT-based 3D Dental Implant Planning

Xinquan Yang, Xuguang Li+7 more|arXiv (Cornell University)|Jan 21, 2026

Dental Radiography and Imaging被引用数 1

ひとこと要約

tldr: RegFreeNet は masked post-implant CBCT データからインプラント位置と傾斜を予測し、術前/術後スキャンの対にならずに学習し、ImplantFairy データセットを公開して大規模評価を可能にする。

ABSTRACT

As the commercial surgical guide design software usually does not support the export of implant position for pre-implantation data, existing methods have to scan the post-implantation data and map the implant to pre-implantation space to get the label of implant position for training. Such a process is time-consuming and heavily relies on the accuracy of registration algorithm. Moreover, not all hospitals have paired CBCT data, limitting the construction of multi-center dataset. Inspired by the way dentists determine the implant position based on the neighboring tooth texture, we found that even if the implant area is masked, it will not affect the determination of the implant position. Therefore, we propose to mask the implants in the post-implantation data so that any CBCT containing the implants can be used as training data. This paradigm enables us to discard the registration process and makes it possible to construct a large-scale multi-center implant dataset. On this basis, we proposes ImplantFairy, a comprehensive, publicly accessible dental implant dataset with voxel-level 3D annotations of 1622 CBCT data. Furthermore, according to the area variation characteristics of the tooth's spatial structure and the slope information of the implant, we designed a slope-aware implant position prediction network. Specifically, a neighboring distance perception (NDP) module is designed to adaptively extract tooth area variation features, and an implant slope prediction branch assists the network in learning more robust features through additional implant supervision information. Extensive experiments conducted on ImplantFairy and two public dataset demonstrate that the proposed RegFreeNet achieves the state-of-the-art performance.

研究の動機と目的

• Eliminate reliance on registration between pre- and post-implant CBCT scans for implant planning.
• Leverage masked post-implant CBCT to enable large-scale multi-center dataset construction.
• Develop a slope-aware, dual-branch network to separately learn implant position and inclination.
• Introduce the Neighboring Distance Perception (NDP) module to recover anatomical context without explicit pre/post supervision.
• Provide publicly accessible ImplantFairy dataset with voxel-level annotations for benchmarking.

提案手法

• Mask implant regions in post-implant CBCT data to enable registration-free learning.
• Use a Neighboring Distance Perception (NDP) module with multi-scale dilated convolutions, a KNet, and Graph Convolution Networks to model tooth-scale anatomical context.
• Employ a dual-branch decoder with a position regression branch and a slope (inclination) prediction branch for explicit geometric regularization.
• Train with a composite loss combining Dice and cross-entropy for segmentation (implant position) and L1 loss for slope estimation.
• Regularize position learning through slope supervision to enforce anatomically plausible implant trajectories.
• Release ImplantFairy, a large-scale implant dataset with 1,622 CBCT scans for benchmarking.

実験結果

リサーチクエスチョン

• RQ1 Can a registration-free framework accurately predict dental implant position and inclination from masked post-implant CBCT data?
• RQ2 How does the NDP module contribute to learning anatomical context without paired pre/post data?
• RQ3 What is the impact of explicitly predicting implant slope as a regularizer on position accuracy?
• RQ4 How does RegFreeNet generalize to external, publicly available CBCT datasets?
• RQ5 What is the value of ImplantFairy for benchmarking registration-free dental implant planning methods?

主な発見

• RegFreeNet achieved a Dice score of 47.51 and IoU of 0.3540 on the SUGH dataset, outperforming all compared models.
• On external datasets, RegFreeNet achieved a Dice score of 33.0 and IoU of 0.2413, showing strong generalization.
• Ablation shows adding NDP and slope prediction branches improves performance, with gains up to 1.87% Dice on SUGH and 8.52% Dice on external data for SPB.
• RegFreeNet with both modules attains the best overall performance among tested methods.
• ImplantFairy is introduced as the first public large-scale dental implant dataset with 1,622 CBCT scans for training and evaluation.