Javascript must be enabled for the correct page display

Syndrome detection using kinship verification

van de Griend, Arianne (2019) Syndrome detection using kinship verification. Bachelor's Thesis, Artificial Intelligence.

[img]
Preview
Text
AI_BA_2019_ArianneVanDeGriend.pdf

Download (437kB) | Preview
[img] Text
toestemming.pdf
Restricted to Registered users only

Download (119kB)

Abstract

Neural developmental disorders are often associated with dysphomic facial features. Currently, dysmorphic features are diagnosed by clinicians observing patients. Computer image analysis can be used to automatically identify these features, but that requires a large dataset for training. This paper investigates the use of a kinship verification classifier to automatically diagnose a rare neural developmental disorder called Koolen-de Vries Syndrome (KdVS). Since it is not possible to build a large enough dataset, this is done using type II errors as a basis for syndrome detection. First, the UB KinFace dataset was split into different train and test sets such that these subsets contained different properties and can be used as train and test set interchangeably. Then, several SVM classifiers were trained on these constructed subsets and evaluated on three tasks: kinship verification, identification, and syndrome detection. The tasks were separated into a genderless and gender-specific version. Unfortunately, no conclusions can be drawn on the applicability of this approach due to shortcomings in the KdVS dataset. We did obtain evidence of a problem with the UB KinFace dataset. There exists a possibility that the dataset contains father-child pairs where the father is not the biological father of the child. This unreliable labeling can be the cause of poor performance on the kinship verification task. In the future computer image processing may be able to aid the diagnosis of patients, however, significant technical limitations need to be overcome.

Item Type: Thesis (Bachelor's Thesis)
Supervisor name: Wiering, M.A.
Degree programme: Artificial Intelligence
Thesis type: Bachelor's Thesis
Language: English
Date Deposited: 17 Jan 2019
Last Modified: 18 Jan 2019 11:37
URI: https://fse.studenttheses.ub.rug.nl/id/eprint/19033

Actions (login required)

View Item View Item