Department of Otolaryngology, Hannover Medical School, Karl-Wiechert-Allee 3, 30625, Hannover, Germany.
Cluster of Excellence Hearing4all, Hannover, Germany.
Biomed Eng Online. 2024 Jul 10;23(1):65. doi: 10.1186/s12938-024-01249-5.
Cochlear implants (CI) are implantable medical devices that enable the perception of sounds and the understanding of speech by electrically stimulating the auditory nerve in case of inner ear damage. The stimulation takes place via an array of electrodes surgically inserted in the cochlea. After CI implantation, cone beam computed tomography (CBCT) is used to evaluate the position of the electrodes. Moreover, CBCT is used in research studies to investigate the relationship between the position of the electrodes and the hearing outcome of CI user. In clinical routine, the estimation of the position of the CI electrodes is done manually, which is very time-consuming.
The aim of this study was to optimize procedures of automatic electrode localization from CBCT data following CI implantation. For this, we analyzed the performance of automatic electrode localization for 150 CBCT data sets of 10 different types of electrode arrays. Our own implementation of the method by Noble and Dawant (Lecture notes in computer science (Including subseries lecture notes in artificial intelligence and lecture notes in bioinformatics), Springer, pp 152-159, 2015. https://doi.org/10.1007/978-3-319-24571-3_19 ) for automated electrode localization served as a benchmark for evaluation. Differences in the detection rate and the localization accuracy across types of electrode arrays were evaluated and errors were classified. Based on this analysis, we developed a strategy to optimize procedures of automatic electrode localization. It was shown that particularly distantly spaced electrodes in combination with a deep insertion can lead to apical-basal confusions in the localization procedure. This confusion prevents electrodes from being detected or assigned correctly, leading to a deterioration in localization accuracy.
We propose an extended cost function for automatic electrode localization methods that prevents double detection of electrodes to avoid apical-basal confusions. This significantly increased the detection rate by 11.15 percent points and improved the overall localization accuracy by 0.53 mm (1.75 voxels). In comparison to other methods, our proposed cost function does not require any prior knowledge about the individual cochlea anatomy.
人工耳蜗是一种可植入的医疗器械，通过对受损内耳的电刺激，使患者能够感知声音并理解言语。刺激通过手术插入耳蜗的电极阵列进行。人工耳蜗植入后，使用锥形束计算机断层扫描 (CBCT) 来评估电极的位置。此外，CBCT 还用于研究电极位置与人工耳蜗使用者听力效果之间的关系。在临床常规中，电极位置的评估是手动进行的，这非常耗时。
本研究旨在优化人工耳蜗植入后从 CBCT 数据中自动定位电极的程序。为此，我们分析了 10 种不同类型电极阵列的 150 个 CBCT 数据集的自动电极定位性能。我们自己实现的 Noble 和 Dawant（计算机科学讲义（包括人工智能和生物信息学讲义分册），Springer，第 152-159 页，2015 年。https://doi.org/10.1007/978-3-319-24571-3_19）的方法作为评估的基准。评估了不同类型电极阵列的检测率和定位精度的差异，并对误差进行了分类。基于此分析，我们开发了一种优化自动电极定位程序的策略。结果表明，特别是在电极间隔较远且插入较深的情况下，可能会导致定位过程中的顶底混淆。这种混淆会阻止电极被正确检测或分配，从而导致定位精度下降。
我们提出了一种用于自动电极定位方法的扩展代价函数，该函数可以防止电极的双重检测，从而避免顶底混淆。这使得检测率提高了 11.15 个百分点，整体定位精度提高了 0.53 毫米（1.75 个体素）。与其他方法相比，我们提出的代价函数不需要任何关于个体耳蜗解剖结构的先验知识。
