University of Twente Student Theses

Login

Identification of pelvic floor features in ultrasound images using an unsupervised machine learning algorithm

Bongers, Twan (2019) Identification of pelvic floor features in ultrasound images using an unsupervised machine learning algorithm.

This is the latest version of this item.

Full text not available from this repository.

Full Text Status:Access to this publication is restricted
Embargo date:1 January 2023
Abstract:Machine learning is pushing itself in our society, and thus also in the medical image analysis. Medical images are currently analyzed by physicians who have had a lot of training and experience. Although these physicians are capable of determining ultrasound images, the physicians can have some assistance from an algorithm. This is, in particular, the case when a physician is required to analyze three-dimensional data. This research is focusing on the analyses of higher-dimensional data of the pelvic floor. Ultrasound creates an accurate representation of the levator ani muscles where the pelvic floor is located. During the capturing of this video (12 and 36 weeks pregnant and 6 months after delivery), the women were asked to perform three movements of contraction: contracted, Valsalva, and rest. The aim is to create an unsupervised neural network, which reduces the high dimensional input data to two data points. These data points can be plotted in a graph to create a scatterplot. Examining this scatterplot creates clusters of similar frames, which are clustered because they have identical input features. Similar images are clustered together which then can be labeled. This labeling can be done by examining the dataset afterward. Analyzing the different movements made by the patients, it is possible to distinguish between the different states of contraction within a single video. These different states of contraction can be used in further research to analyze the maximum contraction or Valsalva. However, the distinguishing between different time frames is difficult, this is because the neural network is trained unsupervised and focuses on the most prominent features available.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:50 technical science in general
Programme:Systems and Control MSc (60359)
Link to this item:http://purl.utwente.nl/essays/79782
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page