Cochlear Implant Fold Detection in Intra-operative CT Using Weakly Supervised Multi-task Deep Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Otolaryngology

Background:

• Cochlear implant (CI) surgery involves inserting an electrode array into the cochlea to restore hearing.
• Electrode array folding during CI surgery can cause cochlear trauma, damage residual hearing, and impair hearing outcomes.
• Intraoperative detection of array folding is crucial for reimplantation but challenging due to reliance on surgeon experience and sometimes unreliable electrophysiological tests.

Purpose of the Study:

• To develop and evaluate an AI-based method for detecting cochlear implant electrode array folding during surgery.
• To address the limitations of current detection methods by utilizing a custom 3D-UNet model trained on synthetic and real CT data.

Main Methods:

• A dataset of CT images was generated, incorporating folded synthetic electrode arrays with realistic metal artifacts.
• A multitask custom 3D-UNet model was trained on this dataset for array fold detection.
• The model's performance was validated on a test set of 207 real post-operative CT scans (7 with folded arrays, 200 without).

Main Results:

• The AI model achieved high accuracy in detecting electrode array folding.
• The model correctly classified all 7 cases of folded arrays.
• Only 3 out of 200 non-folded array cases were misclassified, indicating strong performance.

Conclusions:

• The developed AI model demonstrates significant promise for accurate intraoperative detection of cochlear implant electrode array folding.
• This technology has the potential to reduce surgical complications and improve hearing restoration success rates in cochlear implant recipients.