Back to search

University of Patras

Σύστημα επεξεργασίας, ανάλυσης και ταξινόμησης εικόνων οπτικής μικροσκοπίας από δείγματα βιοψίας παχέος εντέρου με τεχνικές αναγνώρισης προτύπων για διαγνωστικές και εκπαιδευτικές εφαρμογές

Abstract

dc:description

This doctoral dissertation is in the field of Medical Informatics and is more specifically focused on Machine Learning and Image Processing and Analysis. The purpose of this dissertation is the design and implementation of a system for processing, analyzing and classifying optical microscopy images of colon biopsy specimens with pattern recognition techniques for diagnostic and educational applications. The specific decision support system can be embedded in an information system for diagnostic and educational applications. To achieve the purpose of the dissertation, three sub-studies were conducted. A bibliographic research was conducted on the methodologies and the modern medical and pedagogical applications of such systems. On the experimental part of the dissertation, the available data set, on which the study was based, consists of 67 biopsy specimens from different patients. Digital microscopy images were taken at various magnifications (50x, 200x & 400x). The classes in which the sample images were assigned are based on their degree of differentiation, from Low degree (grade I) to High degree of differentiation (grade II and grade III). In the first study, developed in the Matlab environment, after feature extraction and reduction by using statistical methods, texture and color features which could successfully distinguish between the two categories were identified. The efficiency of the classification between Low and High Grade was proved to be high according to the value of the ROC curve. For the second study, an hDL-system was designed to separate Low from High Grade images, using a pre-trained convolutional neural network in order to produce the DL-features with SVM as classifier. In addition, the separation accuracy of the hDL-system was compared with the corresponding supervised machine learning system (sMLsystem). Both systems were implemented in the Matlab environment. The Bootstrap method was used as the evaluation method for both learning systems. The system was designed using the CUDA platform for parallel programming on our Nvidia graphics card microprocessors. The third comparative study, on the one hand, focused on possible differences in the texture and color of the epithelial tissue of patients who had survived five years after initial diagnosis and those of patients who did not survive; on the other hand, explored Machine Learning methods for predicting five-year survival in CRC patients. In conclusion, the supervised Machine Learning system revealed useful information regarding the classification of images based on the grade, as well as the classification between the five-year survival of the patients with high accuracy. The design of the hybrid DL system attained higher accuracies, as compared to the SML system, and proved to be robust and consistent discriminating low-grade from high-grade colorectal cancer (CRC) lesions and assessing the 5-year survival of patients with CRC.

Degree

thesis:*
Grantor dc:publisher
University of Patras
Year dc:date
2021

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Theodosi, Angeliki
  • Θεοδόση, Αγγελική

Subjects

dc:subject × 20

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/51152

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Theodosi, Angeliki; Θεοδόση, Αγγελική. Σύστημα επεξεργασίας, ανάλυσης και ταξινόμησης εικόνων οπτικής μικροσκοπίας από δείγματα βιοψίας παχέος εντέρου με τεχνικές αναγνώρισης προτύπων για διαγνωστικές και εκπαιδευτικές εφαρμογές. University of Patras, 2021. http://hdl.handle.net/10442/hedi/51152