EARSeg3D: An Automated Segmentation Method for Ear Canal Impressions

Abstract

Hearing impairment is a significant global public health issue. Hearing aids, as critical devices for improving the auditory experience of individuals with hearing loss, require personalized customization that decisively affects wearing comfort and acoustic performance. Traditional ear canal impression fabrication relies on manual procedures, which are time-consuming, cause patient discomfort, and introduce subjective errors. Although deep learning has achieved remarkable progress in medical image analysis, its application in the core steps of hearing aid customization—namely, automated semantic understanding and three-dimensional structural analysis of ear canal impressions—remains limited. To address this research gap, this paper proposes EARSeg3D, an end-to-end semantic segmentation method for ear canal impression point clouds based on PointNet++. Targeting the complex geometric characteristics of ear canal morphology, the method introduces a Local Context Fusion Module (LCFM) built upon PointNet++’s hierarchical feature learning framework. By encoding relative spatial positional relationships and feature gradients among points within local neighborhoods, LCFM enhances the network’s ability to perceive the intricate geometric topology of the ear canal. Following a six-class semantic labeling standard, this study partitions ear canal impression point clouds into six semantic regions: ear canal tip, middle ear canal, ear canal base, redundant material, concha cavity, and concha cymba. Ablation study results on a self-constructed dataset show that with the LCFM module, the model achieves a mean Intersection over Union (mIoU) of 92.31% and an overall accuracy (OA) of 83.36%, which are 1.18 and 1.07 percentage points higher than the baseline model, respectively. Visualization results further demonstrate that the method effectively mitigates jagged segmentation boundaries, achieving smoother and more precise segmentation in curved regions of the ear canal and at the junctions between different anatomical structures.