University of Twente Student Theses

Login

Catch the A-train : Detection of A-trains in EMG during vestibular schwannoma resection making use of Neural Networks

Huppelschoten, M. (2019) Catch the A-train : Detection of A-trains in EMG during vestibular schwannoma resection making use of Neural Networks.

[img] PDF
4MB
Abstract:Introduction: Vestibular schwannomas (VS) comprise 75% to 86% of the tumours located in the cerebellopontine angle (CPA). During resection surgery on VS facial nerve damage may occur. In this study an attempt is made to detect A-train time, a suggested predictor for facial nerve damage. This is done by making use of neural networks. Methods: EMG data of 38 patients is available, data of 11 patients is annotated. A convolutional neural network (CNN) and a U-net are trained on datasets with different structures. The model created by the best performing network is analysed and used to predict A-train times of all patients. A stastical analysis is performed, by using the Spearman's rho, to determine the correlation between the found A-train times and the post-op House Brackmann (HB) score, a score to assess facial nerve damage. Results: The U-net performed best on all datasets and is used to create a model. This model has a sensitivity of 92.4% and a specificity of over 99.9% for detecting A-trains in all annotated patients. No correlation was found between the detected A-train times and the HB score directly post-op, 6 weeks, 6 months and 1 year post-op (Spearman's rho of -0.2682 (p=0.1035), -0.1703 (p=0.3136), -0.0301 (p=0.8840), -0.1438 (p=0.5569), respectively). Additionally, no correlation was found between detected A-train time and ΔHB, the difference between pre-op HB score and the last determined HB score (-0.1054 (p=0.5288)). Conclusion: This study shows that making use of neural networks is a promising technique in enabling real-time pattern detection in EMG data. Because no correlation between A-train time and HB score was found, the A-train time is not considered to be a predictor of facial nerve damage. More research is needed to find a predictor for post-op facial nerve damage and towards the etiology of A-trains to further understand their clinical relevance.
Item Type:Essay (Master)
Clients:
RadboudUMC, Nijmegen, Netherlands
Faculty:TNW: Science and Technology
Subject:44 medicine
Programme:Technical Medicine MSc (60033)
Link to this item:http://purl.utwente.nl/essays/79796
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page