Μελέτη γνωστών αλγορίθμων εκμάθησης γράφων, για την αντιμετώπιση του καρκίνου του τραχήλου της μήτρας
KeywordsΚαρκίνος τραχήλου μήτρας ; HPV ; Pap test ; HPV DNA test ; mRNA test ; Flow test ; P16 test ; Δομημένο πιθανοτικό μοντέλο ; Μπεϋζιανά δίκτυα ; Συστήματα υποστήριξης κλινικής απόφασης ; Εξόρυξη γνώσης ; Αλγόριθμος K2 ; Γενετικοί αλγόριθμοι ; Αλγόριθμος Hill climbing ; Αλγόριθμος Simulated annealing ; Cervical cancer ; Built probabilistic model ; Bayesian networks ; Clinical decision support systems ; Mining ; Algorithm K2 ; Genetic algorithms ; Algorithm Hill climbing ; Simulated annealing algorithm
The present diploma thesis deals with the comparison of well known algorithms which classify test results for the diagnosis of cervical cancer, to provide the best possible recommendations for the treatment of this type of cancer. This particular cancer is very common, especially in women who are not examined proactively, however the causes for its appearance have not yet determined with accuracy and validity. An important step towards the investigation of what causes this particular type of cancer was its correlation with different types of HPV virus. Still the Pap test, which was invented in 1943 from Greek doctor Georgios Papanikolaou, is a big milestone in the proactive examination and treatment. Because of this test, the mortality rate from cervical cancer decreased to 74%, which is indicative of the importance of the invention. However, the diagnosis of disease through the Pap test is not 100% valid and successful. The fault prediction rate especially for the squamous cells of undetermined significance (ASCUS - atypical squamous cells of undetermined significance) needs further investigation. To ensure correct diagnosis, it is necessary to add new techniques for detecting cervical lesions, such as HPV DNA test, mRna test, the flow test and the P16 test. Therefore, in case the Pap test of a patient is positive, the patient should be submitted to additional tests in order to investigate exactly what is happening. In our analysis we used data from 380 women who had made all 5 tests above. The diagnostic conclusion from comparing the results of the algorithms could be used as an indicator for the clinician. In this case, the Bayesian networks function as part of clinical decision support systems, and are preferred in the case of diagnosis as well as managing a better degree the element of uncertainty. Specifically, the work focused on the effort of clarification of cases had resulted ASCUS Pap smear to obtain the doctor better picture of the health of the test.