Manual Segmentation Errors in Medical Imaging. Proposing a Reliable Gold Standard
Manual segmentation is ubiquitous in modern medical imaging. It is a tedious and time-consuming process that is also operator-dependent and due to its low reproducibility, presents to specialist a challenge to reach consensus when diagnosing from an image. In the diagnosis of several abnormalities, geometrical features such as distances, curvatures, volumes, areas, and shapes are used to derive verdicts. These features are only quantifiable if the measuring structures can be separated from other elements in the image. The process of manual segmentation provides the analysis with a response to the question of the limits, and those limits are not easy to identify. Despite all the mentioned drawbacks, manual segmentation is still used in medical imaging analysis or employed to validate automatic or semi-automatic methods. Intending to quantify the operator variability of the process, we have created a controlled environment and run segmentations on known volumes scanned with Magnetic Resonance. The strategy proposed here suggests a mechanism to establish gold standards for geometrical readings in medical imaging; thus measuring instruments can be analyzed and certified for the task.
KeywordsSegmentation standards Medical image quantification Repeatable segmentations
- 4.Fenster, A., Chiu, B.: Evaluation of segmentation algorithms for medical imaging. IEEE EMB, January 2006Google Scholar
- 9.Moeskops, P., Viergever, M.A., Mendrik, A.M., de Vries, L.S., Benders, M.J.N.L., Išgum, I.: Automatic segmentation of MR brain images with a convolutional neural network. Fuel 9–14 (2014) Google Scholar
- 10.Pohle, R., Toennies, K.D.: Segmentation of medical images using adaptive region growing. In: Proceedings of Medical Imaging, vol. 4322, July 2001Google Scholar
- 13.Yan, P., Kassim, A.A.: Medical image segmentation using minimal path deformable models with implicit shape priors. In: IEEE EMB, October 2006Google Scholar