Exploration of automated measurement for ossicular chains based on 3-dimensional geometric information

The ossicular chain (malleus, incus, and stapes) is the key middle-ear structure responsible for sound transmission, and abnormalities in any component can disrupt conduction and cause conductive hearing loss. Middle-ear diseases such as otitis media, cholesteatoma, and trauma frequently compromise ossicular integrity, while ossicular-chain reconstruction can effectively restore sound transmission and improve hearing. Because personalized prosthesis preparation depends on accurate preoperative morphometric assessment, obtaining reliable quantitative ossicular parameters is fundamental for surgical planning. From an imaging standpoint, computed tomography is widely used to evaluate middle-ear lesions, yet conventional high-resolution CT often fails to fully visualize delicate structures such as the stapes due to limited spatial resolution. Ultra-high-resolution CT (U-HRCT) markedly improves temporal-bone visualization and provides high-quality data that better support precise segmentation of fine anatomy. Nevertheless, even with U-HRCT, ossicular measurement in clinical workflows still largely relies on manual operations that are time-consuming and labor-intensive, making large-scale quantitative analysis impractical. Existing morphometric studies often use calipers or microscopy on cadaveric specimens or intraoperative measurements, which are detailed but unsuitable for in vivo preoperative assessment. Alternatively, manual or semiautomatic measurements on 2D CT images remain burdensome and frequently require repeated measurements for reliability; their accuracy depends heavily on radiologist expertise. In routine practice, variability in clinician experience and the lack of standardized training and unified measurement criteria can further lead to inconsistent results. “Accordingly, automated ossicular measurement is expected to improve accuracy, repeatability, and efficiency while reducing radiologists’ workload, but it remains underexplored. Prior template/atlas-based approaches may depend on an “average” template and substantial data for template construction, can be less reliable for anatomically variant cases, and have not fully addressed stapes measurements, leaving detailed in vivo ossicular morphometrics relatively scarce.” said the author Mengshi Zhang, a researcher at Beijing Friendship Hospital, “To address this gap, we proposes and validates an automated U-HRCT-based pipeline that couples automated segmentation with 3D geometric information and a surrounding-box strategy to automatically quantify 12 ossicular-chain parameters.”

This study targets the ossicular chain (malleus–incus–stapes), whose abnormalities can cause conductive hearing loss. Because ossiculoplasty requires accurate preoperative measurements for prosthesis planning, but current cadaver-based or manual 2D measurements are laborious and expertise-dependent, the authors propose a systematic pipeline for automated segmentation and measurement on ultra-high-resolution CT (U-HRCT). For data and methods, the study includes 140 patients (226 normal ears) and defines 12 measurement parameters, validated against manual measurements. Segmentation uses TransUnet with multi-view fusion (coronal/sagittal/axial) and an active-contour loss to improve stapes integrity. After 3D reconstruction/smoothing, geometric feature points are detected (e.g., via curvature analysis and bounding-box constraints) to compute distances, angles, and volumes automatically.

Segmentation results indicated that the malleus and incus could be completely separated in all 226 ears, whereas complete segmentation of the malleus, incus, and stapes was achieved in only 47 ears, underscoring the stapes as the main difficulty. For measurement validation, the same observer performed three repeated manual measurements of eight parameters, yielding ICC values greater than 0.8 and indicating high intraobserver consistency. In the 47-ear subset with complete segmentation, automated and manual measurements differed in no parameter (all P > 0.05), and inter-method agreement was high (all ICC > 0.75), supporting the comparability of automated measurements to manual ones. For automated morphometric outputs, malleus and incus parameters were analyzed in all 226 ears, whereas stapes-related parameters and the incudostapedial joint angle were analyzed only in the 47-ear subset. Most parameters exhibited significant sex-based differences, but stapes footplate length, incudostapedial joint angle, and stapes volume did not differ between sexes (P > 0.05). When further stratified by laterality within sex, significant right–left differences were observed only for incus total height in females (P = 0.017) and malleus volume in males (P = 0.037). Across three adult age groups (18–30, 31–60, 61–90), no significant differences were found for any parameter except malleus and incus volumes (P = 0.015 and P = 0.031), suggesting overall stability of adult ossicular morphology.

In the concluding discussion, the authors emphasize that their proposed automated pipeline can translate U-HRCT–based 3D reconstructions of the ossicular chain into reproducible quantitative metrics. Automated measurements were reported to closely match manual measurements, indicating the potential to replace manual work in clinical workflows and thereby improve efficiency and consistency. The paper further highlights the clinical relevance of ossicular morphometry for disease diagnosis, preoperative planning, and individualized prosthesis preparation, arguing that automated 3D measurement can provide data support for more precise and personalized surgical decision-making. At the same time, the authors note important limitations. Measurement reliability is constrained by segmentation quality, and this dependency becomes critical for anatomically complex structures such as the stapes: incomplete stapes segmentation can hinder key feature-point selection and downstream parameter computation. Moreover, because the training annotations and their review depend on expert experience, annotation variability may indirectly affect automated segmentation performance. “Accordingly, future work is oriented toward improving stapes segmentation accuracy to provide a stronger foundation for subsequent automated measurement and medical diagnosis.” said Mengshi Zhang.

Authors of the paper include Mengshi Zhang, Yufan Zhang, Sihui Guo, Xiaoguang Li, Li Zhuo, Yuxue Ren, Wei Chen, Yili Feng, Ruowei Tang, Han Lv, Pengfei Zhao, Zhenchang Wang, and Hongxia Yin.

This work was supported by the National Key R&D Program of China (2020YFA0712203), National Natural Science Foundation of China (grant nos. 62371316, 82302282, and 62276012), Beijing Science and Technology Plan Project (grant no. Z241100009024020), Beijing Scholar 2015 (grant no. [2015]160), and Capital’s Funds for Health Improvement and Research (grant no. 2022-1-1111).

The paper, “Exploration of Automated Measurement for Ossicular Chains Based on 3-Dimensional Geometric Information” was published in the journal Cyborg and Bionic Systems on Jul, 2, 2025, at DOI: 10.34133/cbsystems.0305.