Back to search

National Technical University of Athens (NTUA)

Ανάπτυξη αλγορίθμων υπολογιστικής νοημοσύνης και ευφυών συστημάτων εκτίμησης κινδύνου και υποστήριξης κλινικής απόφασης για την έγκαιρη διάγνωση και την εξατομικευμένη διαχείριση γυναικών με ενδοεπιθηλιακές αλλοιώσεις τραχήλου μήτρας

Abstract

dc:description

Cervical cancer (CxCa) is the second most common cancer and the second leading cause of cancer death worldwide in women between 15 and 44 years old. Persistent infection by oncogenic type of Human Papilloma virus (HPV) is considered the main causative agent of CxCa. HPV infection is the commonest sexually transmitted infection worldwide, however in 90% of the cases the infection automatically regresses. Pap test has been the most valuable tool for CxCa prevention for the last 60 years. However, the interpretation of Pap smears is subjective and is prone to human error. As a result, many CxCa cases are due to underestimated abnormalities present in cytology. Today, a variety of molecular tests and biomarkers are available, that aim to improve CxCa prevention, either by replacing Pap test or by supporting the cytological diagnosis. The most important of these are based on the detection of HPV (HPV DNA testing), the detection of E6/E7 mRNA expression of high risk HPV types (HPV E6/E7 mRNA testing), as well as immunocytochemistry techniques, such as the overexpression of p16INK4a. However, each of the available methods presents its own performance, advantages and limitations, while all are either highly sensitive or highly specific, but not both at the same time. Moreover, efforts to combine them have not so far led to the desired results. Therefore, even today, there is no perfect method for efficient CxCa screening and effective management of women with abnormal screening results. A significant percentage of low-grade lesions in cytology (ASCUS or LSIL) harbours high-grade cervical intraepithelial neoplasia (CIN2/3). These women are at true risk of developing CxCa. Women with low-grade lesions are monitored strictly, as the majority of these lesions automatically regress, while women with high-grade lesions or cancer (CIN2+) require immediate treatment. Today, the widely accepted triage options of ASCUS/LSIL are either immediate colposcopy, or cytological surveillance by frequent Pap tests. The policy of immediate referral to colposcopy can result in colposcopy clinics’ overload, as well as over-intervention and/or over-treatment in the case of subtle colposcopic findings. On the other hand, repeated Pap tests’ policy incorporates the risk of missing high-grade lesions, increases the psychological burden of women, increases non-conformance rates as well as CxCa organised screening programs’ cost. The scope of this doctorate research is the development of artificial intelligence algorithms and intelligent risk assessment and clinical decision support systems (CDSS) for the personalised management of women with cervical pathology and the early and accurate detection of those being at true risk of developing CxCa. The presented work was based on a referral population of 740 women. For each case, the results of five of the most important and widely used CxCa tests (ThinPrep Pap test, HPV DNA full genotyping test, HPV E6/E7 mRNA 5-type genotyping test, HR-HPV E6/E7 mRNA test and p16INK4a) were available, along with the histology result. The data has been analysed and processed utilizing a plethora of computational intelligent algorithms, such as artificial neural networks, Bayesian networks, random forests, genetic algorithms, as well as advanced feature selection and data mining techniques. This work led to the development of several intelligent models, which, by combining the results of the various tests, provide clinicians with patient-specific predictions of the diagnostic outcome and identify women at true risk of developing CxCa. By combining the results of the five aforementioned tests, the Neural Network CDSS presented significantly higher performance in comparison to the medical tests for the detection of CIN2+, while it managed to correctly identify the actual histology of most of the cases. The Weighted Random Forests CDSS may give results for cases with missing examinations’ results and may provide risk estimations for the presence of cervical intraepithelial lesions or the development of CxCa. Moreover, a novel framework of hybrid genetic algorithms has been proposed, namely Genetic Algorithm Matrices (GAM), which aims at finding the global optimal solution of a classification problem by simultaneously applying model selection, feature selection and parameter optimisation. Adopting a set of eight GA matrices, an optimal classification model was extracted, presenting at the same time high sensitivity (88.9%), high specificity (93.5%), high positive predictive value (80%) and high negative predictive value (96.8%) for the detection of CIN2+. The high and balanced performance of this model, in conjunction with the fact that it is able to estimate risk assessments for CIN2+ and to deal with cases with missing data, make it a system that can be used for the optimal personalised management of women with positive tests’ results in screening. Furthermore, the usefulness of utilising lifestyle data has been explored. Using the GAM framework, an intelligent model was developed which combines cytology, HPV DNA and lifestyle data. This model presented better performance for the detection of CIN2+ in comparison with cytology and HPV DNA testing. Besides the development of the intelligent models, by analysing the available data utilising data mining techniques, we were able to produce new knowledge regarding the correlation and the efficacy of several medical tests in detecting CIN2+. By applying Bayesian inference, several combinatorial approaches of the tests were explored, resulting to the identification of an optimal strategy for the management of women with abnormal screening results. This work has also led to the introduction of a new evaluation measure: the robustness of the tests. Finally, a web-based platform has been developed, which, by incorporating the intelligent models, serves as an integrated web-based CDSS for CxCa (CxCaDSS). By providing risk assessments, the developed intelligent models may support clinicians for the early detection, the efficient monitoring and personalised management of women who are at risk of harbouring, developing or relapsing CIN2+. In this way, the presented models aim to improve CxCa prevention, to support monitoring and treatment of women with CIN2+, as well as to support decision making during post-treatment follow-up.

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2017

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Bountris, Panagiotis
  • Μπούντρης, Παναγιώτης

Subjects

dc:subject × 42

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/42047

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

Bountris, Panagiotis; Μπούντρης, Παναγιώτης. Ανάπτυξη αλγορίθμων υπολογιστικής νοημοσύνης και ευφυών συστημάτων εκτίμησης κινδύνου και υποστήριξης κλινικής απόφασης για την έγκαιρη διάγνωση και την εξατομικευμένη διαχείριση γυναικών με ενδοεπιθηλιακές αλλοιώσεις τραχήλου μήτρας. National Technical University of Athens (NTUA), 2017. http://hdl.handle.net/10442/hedi/42047