Artikel
Deep learning-assisted classification of shunt valves on X-ray images
Automatische Erkennung von Shuntventilen anhand von Röntgenbildern mit Hilfe von Deep Learning
Suche in Medline nach
Autoren
-
Thomas Rhomberg
- Inselspital, Universitätsspital Bern, Bern, Schweiz
-
Arsany Hakim
- Inselspital, Universitätsspital Bern, Bern, Schweiz
-
Michael Murek
- Inselspital, Universitätsspital Bern, Bern, Schweiz
| Veröffentlicht: | 25. Mai 2022 |
|---|
Gliederung
Text
Objective: Over the recent decades, the number of different manufacturers and models of cerebrospinal fluid shunt valves constantly increased. Proper identification of shunt valves on x-ray images is crucial to neurosurgeons and radiologists to derive further details of a specific shunt valve, such as opening pressure settings and MR scanning conditions. The main aim of this pilot study is to evaluate the feasibility of an AI-assisted shunt valve detection system.
Methods: In this pilot study, the dataset contained 829 anonymized images of four different commonly used shunt valve types from neurosurgical patients collected from one institutional PACS. All images were acquired from skull x-rays or scout CT images and were cropped in a random size containing the shunt valve. In detail, the dataset contained images of 130 Miethke proGAV 1, 286 Miethke proGAV 2, 226 Codman Hakim Programmable, and 187 Medtronic PS Strata valves. An implementation in Python with the FastAi library [1] was used to resize all images to 460x460 pixels and to train a convolutional neural network with 101 layers using 80% of the images as a training set and 20% of the images for validation.
Results: Overall, our model achieved an F1-score of 0.97 to predict the correct shunt valve model. Breaking down the performance metric for each shunt valve, we could achieve an F1-score of 0.97 for the Codman Hakim Programmable, 0.99 for the Medtronic PS Strata, 0.95 for the Miethke proGAV 1, and 0.99 for the Miethke proGAV 2 shunt valve.
Conclusion: This technology has the potential to automatically detect different shunt valve models in a fast and precise way and may facilitate the identification of an unknown shunt valve on X-ray or CT scout images. Our results are in line with recent publications [2] demonstrating the feasibility of AI-assisted shunt valve identification. The current model is able to distinguish between four different shunt valve types. A larger dataset including the most common shunt valve models is required to make proper use of this technology in daily routine.
