University of Thessaly (UTH)
Computational signal analysis methods for information extraction with applications in biomedicine
Abstract
dc:descriptionThis doctoral thesis explores computational approaches for signal analysis and information extraction. In view of scientific challenges for developing innovative solutions with a broad social impact, it investigates various applications in biomedicine. In this context, a novel signal analysis methodology, based on fuzzy logic, was proposed. This methodology, called Fuzzy Phrases, expresses signals or feature representations of signals using sets of words. Each of these words is represented by a fuzzy set. The words form phrases, which are obtained by the aggregation of the fuzzy sets. Experiments on publicly available datasets showed that it outperforms relevant state-of-the-art approaches. Advantages of the proposed signal representation approach include intuitiveness and tolerance to uncertainty from imprecise or missing information. A significant part of this thesis is devoted to the analysis of endoscopic images, in the context of gastrointestinal tract examination. Endoscopic examination using a wireless camera (Wireless Capsule Endoscopy or WCE) has been established to study and diagnose gastrointestinal pathological conditions. The examination of the patient is performed by a swallowable camera that has the size of a large vitamin pill. The examination with WCE is a non-invasive screening procedure. During the examination, the wireless camera captures and transmits thousands of images of the gastrointestinal tract of the patient. Significant drawback of this examination is the large volume of images received; resulting in the capture of thousands of color images (> 100,000 images per patient). The clinical diagnosis of health conditions requires the review of this large number of collected images by medical specialists. Usually, the reviewers reach their human limits by trying to maintain their concentration undistracted in order to examine this large number of images within an average of 60-90min. This explains why this examination is prone to human errors and has low diagnostic accuracy. To address this important problem, image analysis methods for information extraction to assist in the clinical diagnosis of pathological conditions, were investigated. The development of such methods aims to support medical decision making by reducing the required human effort and by providing a second opinion on the examined medical problem. In-depth research of state-of-the-art methods of image processing and analysis in the literature led to several contributions. One of the contributions is the detailed recording of the literature on technological developments in the field of wireless capsule endoscopy in the last five years 2013-2018. Supervised classification methods were developed requiring only weak annotation of images by experts, i.e., the experts do not need to annotate in detail the areas of abnormality. Experimental results proved the suitability of the proposed weakly supervised methods for the detection of images with pathological content. Subsequently, algorithms were developed to identify the areas of abnormalities within these images. As the results from weak annotation approaches were encouraging for the classification between abnormal and normal images, supervised methods were adopted also for the semantic interpretation of the whole content of images using multi-label classification methods. The effort to interpret the whole content of endoscopic images proved to be reasonable, as the results for image classification and pathology finding were further improved compared to previous approaches.An important aspect in image analysis is the detection of interest points. Such points are useful for the extraction of features from the areas around them. The development of these methods has the effect of limiting the sampling points of the images to these points only, and thus simplifying the computational cost of the analysis process. To this end, during this doctoral research the algorithms suggested by the literature for the detection of points of interest in images were studied and a novel detection algorithm for endoscopic images was proposed. The proposed algorithm was based on the color characteristics of the images. The experimental results were satisfactory with respect to locating points within pathological areas, and it is important to note that the algorithm was unsupervised, as it has no requirement of training, based on previous knowledge in order to detect points. The work presented in this thesis provides the basis for further research on the topics studied. Particularly in the context of capsule endoscopy, it has contributed methods that can be practically used by physicians to detect various gastrointestinal abnormalities in endoscopic images.
Degree
thesis:*- Grantor dc:publisher
- University of Thessaly (UTH)
- Year dc:date
- 2020
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Vasilakakis, Michael
- Βασιλακάκης, Μιχαήλ
Subjects
dc:subject × 22- Ανάλυση σήματος
- Ψηφιακή ανάληση εικόνας
- Ασαφή σύνολα και ασαφή λογική
- Ανάλυση ενδοσκοπικών εικόνων ασύρματης ενδοσκοπικής κάψουλας
- Ταξινόμηση ασθενώς επισημειωμένων εικόνων
- Εντοπισμός σημείων ενδιαφέροντος σε ενδοσκοπικές εικόνες
- Ταξινόμηση πολλαπλών ετικετών
- Signal analysis
- Digital image analysis
- Fuzzy sets and fuzzy logic
- Wireless capsule endoscopy image analysis
- Weakly annotated image classification
- Interest point detection in endoscopic images
- Multi-label classification
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Natural Sciences
- Computer and Information Sciences
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
dc:rights- Statement dc:rights
-
- BY_NC_ND
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/47314