No description available.

Public transport plays a vital role in society as the economy, the degree of sustainability and the qualityof life of a city is directly affected by transportation. A shift in modal share towards public transport isassociated with many benefits such as increased air quality and improved space allocation within thecity. To further promote public transport, an appropriate measure of competitiveness is required toevaluate the impact of past and future transport policies. This study introduces the journeys per capitaratio as a new way of measuring public transport competitiveness. Firstly, the key factors affecting thepublic transportation usage rate expressed as the journeys per capita ratio are identified to evaluatethe impact of public transport provider efforts. For this purpose, data for a total of 32 explanatoryvariables and a scope of 218 regions for seven consecutive time frames are collected. Secondly,geographically weighted regression (GWR) – a local regression-based spatial analysis technique – isperformed to test if the journeys per capita ratio is a suitable target variable to predict the impact ofcertain transport supply changes. A traditional global ordinary least square (OLS) model is conductedas well to compare if a local model could be more beneficial. The GWR and the OLS model are trainedwith the data of previous years and tested with data from the consecutive following years. Thirdly,further temporal and socio-economic based cluster analyses are performed to assess the validity andthe explanatory power of the journeys per capita ratio. The conducted analyses reveal that thejourneys per capita ratio is a superior measure for assessing public transport competitiveness.Goodness of fit statistics and estimation results demonstrate that the GWR model has betterprediction accuracy and is more capable of retrospectively predicting the impact of previous transportpolicies. / Kollektivtrafiken har en avgörande roll i samhället då ekonomin, graden av hållbarhet och städerslivskvalité är direkt påverkad av transport. En förändring av transportanvändning från bil motkollektivtrafik är förknippad med flera fördelar, såsom ökad luftkvalitet och förbättrad rumsligallokering inom staden. För att ytterligare främja kollektivtrafik krävs ett lämpligt mått påkonkurrenskraft för att utvärdera effekterna av tidigare och framtida transportpolitik. Den här studienintroducerar resor per capita-förhållanden som ett nytt sätt att mäta kollektivtransportenskonkurrenskraft. För det första identifieras nyckelfaktorerna som påverkar användningsgraden förkollektivtrafik, uttryckt som förhållandet resor per capita för att utvärdera effekten avkollektivtrafikleverantörens insatser. För det här syftet har data för totalt 32 variabler och ett omfångav 218 regioner under sju, på varandra, följande tidsramar har samlats in. För det andra har Geografisktviktad regression (GWR), vilket är en lokal regressionsbaserad rumslig analysteknik, använts för atttesta om resor per capita-förhållanden är en lämplig målvariabel för att förutsäga effekterna av vissatransportförändringar. En traditionell Global ordinary least square model (OLS) har också använts föratt jämföra om en lokal modell är mer fördelaktig. GWR och OLS modellerna har tränats med data fråntidigare år och testats med data från följande år. För det tredje har ytterligare tidsmässigsocioekonomisk baserad klusteranalys utförts för att bedöma validiteten och förklaringsförmågan förresornas förhållande per capita. De genomförda analyserna pekar på att förhållandet resor per capitaär ett fördelaktigt mått för att bedöma kollektivtrafikens konkurrenskraft. Goodness of fit statistics ochde uppskattade resultaten visar att GWR-modellen har en bättre förmåga att göra noggrannaförutsägelser och är mer kapabel att i efterhand förutsäga effekterna av tidigare transportpolitik.

We formulate Local Shape Descriptor selection for model-based object recognition in range data as an optimization problem and offer a platform that facilitates a solution. The goal of object recognition is to identify and localize objects of interest in an image. Recognition is often performed in three phases: point matching, where correspondences are established between points on the 3-D surfaces of the models and the range image; hypothesis generation, where rough alignments are found between the image and the visible models; and pose refinement, where the accuracy of the initial alignments is improved. The overall efficiency and reliability of a recognition system is highly influenced by the effectiveness of the point matching phase. Local Shape Descriptors are used for establishing point correspondences by way of encapsulating local shape, such that similarity between two descriptors indicates geometric similarity between their respective neighbourhoods. We present a generalized platform for constructing local shape descriptors that subsumes a large class of existing methods and allows for tuning descriptors to the geometry of specific models and to sensor characteristics. Our descriptors, termed as Variable-Dimensional Local Shape Descriptors, are constructed as multivariate observations of several local properties and are represented as histograms. The optimal set of properties, which maximizes the performance of a recognition system, depend on the geometry of the objects of interest and the noise characteristics of range image acquisition devices and is selected through pre-processing the models and sample training images. Experimental analysis confirms the superiority of optimized descriptors over generic ones in recognition tasks in LIDAR and dense stereo range images. / Thesis (Ph.D, Electrical & Computer Engineering) -- Queen's University, 2009-09-01 11:07:32.084

In forensic autopsies, accurately estimating the postmortem interval (PMI) is crucial. Traditional methods, relying on physical parameters and police data, often lack precision, particularly after approximately two days have passed since the person's death. New methods are increasingly focusing on analyzing postmortem metabolomics in biological systems, acting as a 'fingerprint' of ongoing processes influenced by internal and external molecules. By carefully analyzing these metabolomic profiles, which span a diverse range of information from events preceding death to postmortem changes, there is potential to provide more accurate estimates of the PMI. The limitation of available real human data has hindered comprehensive investigation until recently. Large-scale metabolomic data collected by the National Board of Forensic Medicine (RMV, Rättsmedicinalverket) presents a unique opportunity for predictive analysis in forensic science, enabling innovative approaches for improving  PMI estimation. However, the metabolomic data appears to be large, complex, and potentially nonlinear, making it difficult to interpret. This underscores the importance of effectively employing machine learning algorithms to manage metabolomic data for the purpose of PMI predictions, the primary focus of this project.  In this study, a dataset consisting of 4,866 human samples and 2,304 metabolites from the RMV was utilized to train a model capable of predicting the PMI. Random Forest (RF) and Artificial Neural Network (ANN) models were then employed for PMI prediction. Furthermore, feature selection and incorporating sex and age into the model were explored to improve the neural network's performance.  This master's thesis shows that ANN consistently outperforms RF in PMI estimation, achieving an R2 of 0.68 and an MAE of 1.51 days compared to RF's R2 of 0.43 and MAE of 2.0 days across the entire PMI-interval. Additionally, feature selection indicates that only 35% of total metabolites are necessary for comparable results with maintained predictive accuracy. Furthermore, Principal Component Analysis (PCA) reveals that these informative metabolites are primarily located within a specific cluster on the first and second principal components (PC), suggesting a need for further research into the biological context of these metabolites.  In conclusion, the dataset has proven valuable for predicting PMI. This indicates significant potential for employing machine learning models in PMI estimation, thereby assisting forensic pathologists in determining the time of death. Notably, the model shows promise in surpassing current methods and filling crucial gaps in the field, representing an important step towards achieving accurate PMI estimations in forensic practice. This project suggests that machine learning will play a central role in assisting with determining time since death in the future.

Ο στόχος αυτής της εργασίας ήταν η αναγνώριση έξι βασικών κινήσεων του χεριού με χρήση δύο συστημάτων. Όντας θέμα διεπιστημονικού επιπέδου έγινε μελέτη της ανατομίας των μυών του πήχη, των βιοσημάτων, της μεθόδου της ηλεκτρομυογραφίας (ΗΜΓ) και μεθόδων αναγνώρισης προτύπων. Παράλληλα, το σήμα περιείχε αρκετό θόρυβο και έπρεπε να αναλυθεί, με χρήση του EMD, να εξαχθούν χαρακτηριστικά αλλά και να μειωθεί η διαστασιμότητά τους, με χρήση των RELIEF και PCA, για βελτίωση του ποσοστού επιτυχίας ταξινόμησης. Στο πρώτο μέρος γίνεται χρήση συστήματος ΗΜΓ της Delsys αρχικά σε ένα άτομο και στη συνέχεια σε έξι άτομα με το κατά μέσο όρο επιτυχημένης ταξινόμησης, για τις έξι αυτές κινήσεις, να αγγίζει ποσοστά άνω του 80%. Το δεύτερο μέρος περιλαμβάνει την κατασκευή αυτόνομου συστήματος ΗΜΓ με χρήση του Arduino μικροελεγκτή, αισθητήρων ΗΜΓ και ηλεκτροδίων, τα οποία είναι τοποθετημένα σε ένα ελαστικό γάντι. Τα αποτελέσματα ταξινόμησης σε αυτή την περίπτωση αγγίζουν το 75%. / The aim of this work was to identify six basic movements of the hand using two systems. Being an interdisciplinary topic, there has been conducted studying in the anatomy of forearm muscles, biosignals, the method of electromyography (EMG) and methods of pattern recognition. Moreover, the signal contained enough noise and had to be analyzed, using EMD, to extract features and to reduce its dimensionality, using RELIEF and PCA, to improve the success rate of classification. The first part uses an EMG system of Delsys initially for an individual and then for six people with the average successful classification, for these six movements at rates of over 80%. The second part involves the construction of an autonomous system EMG using an Arduino microcontroller, EMG sensors and electrodes, which are arranged in an elastic glove. Classification results in this case reached 75% of success.

Artificial Neural Networks (ANNs) have been comprehensively researched, both from a computer scientific perspective and with regard to their use for predictive modelling in a wide variety of applications including hydrology and the environment. Yet their adoption for live, real-time systems remains on the whole sporadic and experimental. A plausible hypothesis is that this may be at least in part due to their treatment heretofore as “black boxes” that implicitly contain something that is unknown, or even unknowable. It is understandable that many of those responsible for delivering Early Warning Systems (EWS) might not wish to take the risk of implementing solutions perceived as containing unknown elements, despite the computational advantages that ANNs offer. This thesis therefore builds on existing efforts to open the box and develop tools and techniques that visualise, analyse and use ANN weights and biases especially from the viewpoint of neural pathways from inputs to outputs of feedforward networks. In so doing, it aims to demonstrate novel approaches to self-improving predictive model construction for both regression and classification problems. This includes Neural Pathway Strength Feature Selection (NPSFS), which uses ensembles of ANNs trained on differing subsets of data and analysis of the learnt weights to infer degrees of relevance of the input features and so build simplified models with reduced input feature sets. Case studies are carried out for prediction of flooding at multiple nodes in urban drainage networks located in three urban catchments in the UK, which demonstrate rapid, accurate prediction of flooding both for regression and classification. Predictive skill is shown to reduce beyond the time of concentration of each sewer node, when actual rainfall is used as input to the models. Further case studies model and predict statutory bacteria count exceedances for bathing water quality compliance at 5 beaches in Southwest England. An illustrative case study using a forest fires dataset from the UCI machine learning repository is also included. Results from these model ensembles generally exhibit improved performance, when compared with single ANN models. Also ensembles with reduced input feature sets, using NPSFS, demonstrate as good or improved performance when compared with the full feature set models. Conclusions are drawn about a new set of tools and techniques, including NPSFS and visualisation techniques for inspection of ANN weights, the adoption of which it is hoped may lead to improved confidence in the use of ANN for live real-time EWS applications.

Two of the most important environmental challenges in the 21st century are to protect the quality of fresh water resources and to utilize renewable energy sources to lower greenhouse gas emissions. This study contributes to the solution of the first challenge by providing methodologies for optimal design of real-time water quality monitoring systems and interpretation of data supplied by the monitoring system to identify potential pollution sources in river networks. In this study, the optimal river water quality monitoring network design aspect of the overall monitoring program is addressed by a novel methodology for the analysis of this problem. In this analysis, the locations of sampling sites are determined such that the contaminant detection time is minimized for the river network while achieving maximum reliability for the monitoring system performance. The data collected from these monitoring stations can be used to identify contamination source locations. This study suggests a methodology that utilizes a classification routine which associates the observations on a contaminant spill with one or more of the candidate spill locations in the river network. This approach consists of a training step followed by a sequential elimination of the candidate spill locations which lead to the identification of potential spill locations. In order to contribute the solution of the second environmental challenge, this study suggests utilizing available excess energy in water distribution systems by providing a methodology for optimal design of energy recovery systems. The energy recovery in water distribution systems is possible by using micro hydroelectric turbines to harvest available excess energy inevitably produced to satisfy consumer demands and to maintain adequate pressures. In this study, an optimization approach for the design of energy recovery systems in water distribution networks is proposed. This methodology is based on finding the best locations for micro hydroelectric plants in the network to recover the excess energy. Due to the unsteady nature of flow in water distribution networks, the proposed methodology also determines optimum operation schedules for the micro turbines.

This thesis deals with machine learning techniques for the extraction of structure and the analysis of the vertebrate olfactory pathway based on related methods. Some of its main contributions are summarized below. We have performed a systematic investigation for classification in biomedical images with the goal of recognizing a material in these images by its texture. This investigation included (i) different measures for evaluating the importance of image descriptors (features), (ii) methods to select a feature set based on these evaluations, and (iii) classification algorithms. Image features were evaluated according to their estimated relevance for the classification task and their redundancy with other features. For this purpose, we proposed a framework for relevance and redundancy measures and, within this framework, we proposed two new measures. These were the value difference metric and the fit criterion. Both measures performed well in comparison with other previously used ones for evaluating features. We also proposed a Hopfield network as a method for feature selection, which in experiments gave one of the best results relative to other previously used approaches. We proposed a genetic algorithm for clustering and tested it on several realworld datasets. This genetic algorithm was novel in several ways, including (i) the use of intra-cluster distance as additional optimization criterion, (ii) an annealing procedure, and (iii) adaptation of mutation rates. As opposed to many conventional clustering algorithms, our optimization framework allowed us to use different cluster validation measures including those which do not rely on cluster centroids. We demonstrated the use of the clustering algorithm experimentally with several cluster validity measures as optimization criteria. We compared the performance of our clustering algorithm to that of the often-used fuzzy c-means algorithm on several standard machine learning datasets from the University of California/Urvine (UCI) and obtained good results. The organization of representations in the brain has been observed at several stages of processing to spatially decompose input from the environment into features that are somehow relevant from a behavioral or perceptual standpoint. For the perception of smells, the analysis of such an organization, however, is not as straightforward because of the missing metric. Some studies report spatial clusters for several combinations of physico-chemical properties in the olfactory bulb at the level of the glomeruli. We performed a systematic study of representations based on a dataset of activity-related images comprising more than 350 odorants and covering the whole spatial array of the first synaptic level in the olfactory system. We found clustered representations for several physico-chemical properties. We compared the relevance of these properties to activations and estimated the size of the coding zones. The results confirmed and extended previous studies on olfactory coding for physico-chemical properties. Particularly of interest was the spatial progression by carbon chain that we found. We discussed our estimates of relevance and coding size in the context of processing strategies. We think that the results obtained in this study could guide the search into olfactory coding primitives and the understanding of the stimulus space. In a second study on representations in the olfactory bulb, we grouped odorants together by perceptual categories, such as floral and fruity. By the application of the same statistical methods as in the previous study, we found clustered zones for these categories. Furthermore, we found that distances between spatial representations were related to perceptual differences in humans as reported in the literature. This was possibly the first time that such an analysis had been done. Apart from pointing towards a spatial decomposition by perceptual dimensions, results indicate that distance relationships between representations could be perceptually meaningful. In a third study, we modeled axon convergence from olfactory receptor neurons to the olfactory bulb. Sensory neurons were stimulated by a set of biologically-relevant odors, which were described by a set of physico-chemical properties that covaried with the neural and glomerular population activity in the olfactory bulb. Convergence was mediated by the covariance between olfactory neurons. In our model, we could replicate the formation of glomeruli and concentration coding as reported in the literature, and further, we found that the spatial relationships between representational zones resulting from our model correlated with reported perceptual differences between odor categories. This shows that natural statistics, including similarity of physico-chemical structure of odorants, can give rise to an ordered arrangement of representations at the olfactory bulb level where the distances between representations are perceptually relevant. / <p>QC 20120224</p>

This thesis deals with the problem of newborn seizre detection from the Electroencephalogram (EEG) signals. The ultimate goal is to design an automated seizure detection system to assist the medical personnel in timely seizure detection. Seizure detection is vital as neurological diseases or dysfunctions in newborn infants are often first manifested by seizure and prolonged seizures can result in impaired neuro-development or even fatality. The EEG has proved superior to clinical examination of newborns in early detection and prognostication of brain dysfunctions. However, long-term newborn EEG signals acquisition is considerably more difficult than that of adults and children. This is because, the number of the electrodes attached to the skin is limited by the size of the head, the newborns EEGs vary from day to day, and the newborns are reluctant of being in the recording situation. Also, the movement of the newborn can create artifact in the recording and as a result strongly affect the electrical seizure recognition. Most of the existing methods for neonates are either time or frequency based, and, therefore, do not consider the non-stationarity nature of the EEG signal. Thus, notwithstanding the plethora of existing methods, this thesis applies the discrete wavelet transform (DWT) to account for the non-stationarity of the EEG signals. First, two methods for seizure detection in neonates are proposed. The detection schemes are based on observing the changing behaviour of a number of statistical quantities of the wavelet coefficients (WC) of the EEG signal at different scales. In the first method, the variance and mean of the WC are considered as a feature set to dassify the EEG data into seizure and non-seizure. The test results give an average seizure detection rate (SDR) of 97.4%. In the second method, the number of zero-crossings, and the average distance between adjacent extrema of the WC of certain scales are extracted to form a feature set. The test obtains an average SDR of 95.2%. The proposed feature sets are both simple to implement, have high detection rate and low false alarm rate. Then, in order to reduce the complexity of the proposed schemes, two optimising methods are used to reduce the number of selected features. First, the mutual information feature selection (MIFS) algorithm is applied to select the optimum feature subset. The results show that an optimal subset of 9 features, provides SDR of 94%. Compared to that of the full feature set, it is clear that the optimal feature set can significantly reduce the system complexity. The drawback of the MIFS algorithm is that it ignores the interaction between features. To overcome this drawback, an alternative algorithm, the mutual information evaluation function (MIEF) is then used. The MIEF evaluates a set of candidate features extracted from the WC to select an informative feature subset. This function is based on the measurement of the information gain and takes into consideration the interaction between features. The performance of the proposed features is evaluated and compared to that of the features obtained using the MIFS algorithm. The MIEF algorithm selected the optimal 10 features resulting an average SDR of 96.3%. It is also shown, an average SDR of 93.5% can be obtained with only 4 features when the MIEF algorithm is used. In comparison with results of the first two methods, it is shown that the optimal feature subsets improve the system performance and significantly reduce the system complexity for implementation purpose.

Σκοπός της παρούσας διδακτορικής διατριβής είναι η ανάπτυξη ενός ολοκληρωμένου συστήματος υποστήριξης της διάγνωσης (Decision Support System - DSS) με χρήση μεθόδων εξόρυξης δεδομένων για την ταξινόμηση επιχρισμάτων βιοψίας με λεπτή βελόνα (Fine Needle Aspiration - FNA). Δύο κατηγορίες επιλέχθηκαν για τα δείγματα FNA: καλοήθεια και κακοήθεια. Το σύστημα αυτό αποτελείται από τις ακόλουθες βαθμίδες: 1) συλλογής δεδομένων, 2) επιλογής δεδομένων, 3) εύρεσης κατάλληλων χαρακτηριστικών, 4) εφαρμογής ταξινόμησης με χρήση μεθόδων εξόρυξης δεδομένων. Επίσης, βασικός στόχος της παρούσας διδακτορικής διατριβής ήταν η βελτίωση της ορθής ταξινόμησης των ύποπτων επιχρισμάτων (suspicious), για τα οποία είναι γνωστή η αδυναμία της μεθόδου FNA να τα ταξινομήσει. Το σύστημα εκπαιδεύτηκε και ελέγχθηκε σε σχέση με το δείγμα για το οποίο είχαμε ιστολογικές επιβεβαιώσεις (ground truth). Για περιπτώσεις οι οποίες χαρακτηρίστηκαν ως μη κακοήθεις από την FNA, και για τις οποίες δεν είχαμε ιστολογικές επιβεβαιώσεις, το δείγμα προέκυψε από την συνεκτίμηση και άλλων κλινικών, εργαστηριακών και απεικονιστικών εξετάσεων. Στα πλαίσια της παρούσας διδακτορικής διατριβής συλλέχθηκαν εξετάσεις FNA θυρεοειδούς από το Εργαστήριο Παθολογοανατομίας του Α’ Τμήματος Παθολογίας της Ιατρικής Σχολής του Πανεπιστημίου Αθηνών. Δεδομένου ότι το εν λόγω εργαστήριο λειτουργεί και σαν κέντρο αναφοράς, σημαντικός αριθμός των δειγμάτων εστάλησαν εκεί και από άλλα Εργαστήρια Παθολογοανατομίας για επανέλεγχο. Το αρχειακό υλικό ήταν πολύ καλά ταξινομημένο σε χρονολογική σειρά αλλά ήταν σε έντυπη μορφή. Αρχικά πραγματοποιήθηκε η ανάλυση απαιτήσεων για τη δομή και το σχεδιασμό της βάσης δεδομένων. Με βάση τα στοιχεία από την τεκμηριωμένη διάγνωση σχεδιάστηκε και αναπτύχθηκε προηγμένο σύστημα για την κωδικοποίηση και αρχικοποίηση των δεδομένων. Με τη βοήθεια του σχεδιασμού και ανάλυσης απαιτήσεων αναπτύχθηκε και υλοποιήθηκε η βάση δεδομένων στην οποία αποθηκεύτηκαν τα δεδομένα προς επεξεργασία. Παράλληλα, με το σχεδιασμό της βάσης έγινε και η προεργασία για το σχεδιασμό και την ανάλυση απαιτήσεων του γραφικού περιβάλλοντος εισαγωγής στοιχείων. Λαμβάνοντας υπόψη ότι το σύστημα θα μπορούσε να χρησιμοποιηθεί και πέρα από τα πλαίσια της παρούσας διδακτορικής διατριβής λήφθηκε μέριμνα ώστε να παρέχεται ένα φιλικό και ευέλικτο προς το χρήστη περιβάλλον. Σύμφωνα με τη μεθοδολογία προσέγγισης η οποία ακολουθήθηκε προηγήθηκε στατιστική ανάλυση των 9.102 συλλεχθέντων δειγμάτων FNA ως προς τα κυτταρολογικά χαρακτηριστικά τους και τις διαγνώσεις. Οι κυτταρολογικές διαγνώσεις των συγκεκριμένων δειγμάτων συσχετίστηκαν με τις ιστολογικές διαγνώσεις, στοχεύοντας στον υπολογισμό της πιθανής επίδρασης και συμβολής κάθε κυτταρολογικού χαρακτηριστικού σε μια ορθή ή ψευδή κυτταρολογική διάγνωση, έτσι ώστε να προσδιοριστούν οι πιθανές πηγές λανθασμένης διάγνωσης. Τα δείγματα τα οποία περιείχαν μόνο αίμα ή πολύ λίγα θυλακειώδη κύτταρα χωρίς κολλοειδές θεωρήθηκαν ανεπαρκή για τη διάγνωση. Οι βιοψίες εκτελέσθηκαν είτε στο Α’ τμήμα του Πανεπιστημίου Αθηνών (οι περισσότερες από τις περιπτώσεις με ψηλαφητούς όζους) είτε αλλού (κυρίως κάτω από την καθοδήγηση του κέντρου αναφοράς). Τα δείγματα επιστρωμένα σε πλακάκια, στάλθηκαν στο κέντρο αναφοράς από διάφορα νοσοκομεία, με διαφορετικά πρωτόκολλα σχετικά με τα κριτήρια εκτέλεσης βιοψίας FNA σε θυρεοειδή. Μετεγχειρητικές ιστολογικές επαληθεύσεις ήταν διαθέσιμες για 266 ασθενείς (κακοήθειες και μη). Το χαμηλό ποσοστό ιστολογικών επαληθεύσεων οφείλεται στην ετερογενή προέλευση των ασθενών και στην έλλειψη ολοκληρωμένης παρακολούθησης και επανελέγχου των ασθενών. Για την αξιολόγηση των δεδομένων χρησιμοποιήθηκαν περιγραφικά στατιστικά μεγέθη όπως, μέση τιμή, τυπική απόκλιση, ποσοστά, μέγιστο και ελάχιστο. Έγιναν επίσης και χ2 δοκιμές επιπέδου σημαντικότητας διαφόρων παραμέτρων για να ελεγχθεί η πιθανή συσχέτιση ή η ανεξαρτησία. Για τη συσχέτιση των κυτταρολογικών και των ιστολογικών διαγνώσεων και την αξιολόγηση των εργαστηριακών ευρημάτων, πέραν των περιγραφικών στατιστικών μεγεθών χρησιμοποιήθηκαν και υπολογισμοί της ευαισθησίας, της ειδικότητας, της συνολικής ακρίβειας, της αρνητικής και θετικής αξίας πρόβλεψης (negative and positive predictive value). Προκειμένου να καθοριστεί εάν μια κατηγορία ασθενειών συσχετίζεται ή όχι με συγκεκριμένες κυτταρολογικές παραμέτρους εφαρμόστηκε μέθοδος ελέγχου στατιστικής σημαντικότητας σε επίπεδο 5% (p < 0,05). Η διαδικασία ακολουθήθηκε για κάθε κατηγορία ασθενειών ή συνδυασμό τους και για κάθε παράμετρο των κυτταρολογικών και αρχιτεκτονικών στοιχείων της κυτταρολογικής διάγνωσης. Τα αποτελέσματα της στατιστικής ανάλυσης επέτρεψαν το διαχωρισμό των δεδομένων σε καλοήθη, κακοήθη, νεοπλασματικά, ύποπτα για κακοήθεια και οριακά με χαρακτηριστικά γνωρίσματα μεταξύ ενός καλοήθους και ενός νεοπλασματικού. Στην συνέχεια αναπτύχθηκε σύστημα υποστήριξης της διάγνωσης χρησιμοποιώντας εξειδικευμένες μεθόδους εξόρυξης δεδομένων. Το σύστημα αποτελείται από τέσσερις βαθμίδες. Η πρώτη βαθμίδα αυτού του συστήματος είναι το περιβάλλον Συλλογής Δεδομένων στην οποία τα δεδομένα αποθηκεύονται στη βάση δεδομένων. Η Δεύτερη Βαθμίδα αυτού του συστήματος αφορά στην Επιλογή Δεδομένων. Σύμφωνα με την καταγραφή των απαιτήσεων, την εισαγωγή και τη ψηφιοποίηση των στοιχείων, δημιουργήθηκαν 111 χαρακτηριστικά για κάθε ασθενή (record). Τα περισσότερα χαρακτηριστικά είχαν τιμές δυαδικού τύπου, αποτυπώνοντας την ύπαρξη ή μη του κάθε χαρακτηριστικού, ενώ κάποιες άλλες είχαν τιμές τύπων αριθμών ή αλφαριθμητικών χαρακτήρων. Από τα 111 χαρακτηριστικά επιλέχθηκαν 60 χαρακτηριστικά τα οποία περιγράφουν τη δομή των επιχρισμάτων ενώ δημιουργήθηκαν άλλα 7 χαρακτηριστικά τα οποία αφορούσαν στην ομαδοποίηση άλλων χαρακτηριστικών. Η Τρίτη Βαθμίδα του συστήματος αφορά στην εύρεση των Κατάλληλων Χαρακτηριστικών. Λόγω του αρχικά υψηλού αριθμού χαρακτηριστικών παραμέτρων (67 ανά περίπτωση), ήταν απαραίτητο να εξαλειφθούν οι χαρακτηριστικές παράμετροι που συσχετίζονταν γραμμικά ή δεν είχαν καμία διαγνωστική πληροφορία. H μέθοδος επιλογής χαρακτηριστικών εφαρμόστηκε πριν από την ταξινόμηση, με γνώμονα την ανεύρεση ενός υποσυνόλου των χαρακτηριστικών παραμέτρων που βελτιστοποιούν σε ακρίβεια τη διαδικασία ταξινόμησης. Εφαρμόστηκε η τεχνική επιπλέουσας πρόσθιας ακολουθιακά μεταβαλλόμενης επιλογής (SFFS). Ο αριθμός των δειγμάτων που χρησιμοποιήθηκαν είναι 2.036 (1.886 καλοήθειες και 150 κακοήθειες). Εξ αυτών, όλες οι κακοήθειες είναι ιστολογικά επιβεβαιωμένες. Επίσης, 140 καλοήθειες είναι ιστολογικά επιβεβαιωμένες με επάρκεια υλικού. Οι υπόλοιπες 1.726 καλοήθειες είναι επιβεβαιωμένες με συνεκτίμηση κλινικών, εργαστηριακών και απεικονιστικών ιατρικών εξετάσεων (υπέρηχοι κ.λπ.). Από τα 2.036 δείγματα, το 25% χρησιμοποιήθηκε για την επιλογή χαρακτηριστικών παραμέτρων, δηλαδή 37 περιπτώσεις κακοήθειας (Malignant) και 472 περιπτώσεις καλοήθειας (Non Malignant). Από την εφαρμογή της τεχνικής (SFFS) επιλέχθηκαν τελικά 12 χαρακτηριστικά ως βέλτιστα για την ταξινόμηση των δεδομένων FNA σε καλοήθη και κακοήθη. Η Τέταρτη βαθμίδα επεξεργασίας είναι η Εφαρμογής Ταξινόμησης με χρήση Μεθόδων Εξόρυξης Δεδομένων ή Ταξινομητής. Για το σκοπό αυτό, επιλέχθηκε να εφαρμοστεί μια πληθώρα αξιόπιστων, καλά επιβεβαιωμένων και σύγχρονων μεθόδων εξόρυξης δεδομένων. Το σύστημα εκπαιδεύτηκε και ελέγχθηκε σε σχέση με το δείγμα για το οποίο είχαμε ιστολογικές επιβεβαιώσεις (ground truth). Η ανεξάρτητη εφαρμογή τεσσάρων αξιόπιστων μεθόδων, Δέντρων Αποφάσεων (Decision Trees), Τεχνιτών Νευρωνικών Δικτύων (Artificial Neural Network), Μηχανών Στήριξης Διανυσμάτων (Support Vector Machine), και Κ - κοντινότερου γείτονα (k-NN), έδωσε αποτελέσματα συγκρίσιμα με αυτά της FNA μεθόδου. Περαιτέρω βελτίωση των αποτελεσμάτων επιτεύχθηκε με την εφαρμογή της μεθόδου πλειοψηφικού κανόνα (Majority Vote - CMV) συνδυάζοντας τα αποτελέσματα από την εφαρμογή των τριών καλύτερων αλγορίθμων, ήτοι των Νευρωνικών Δικτύων, Μηχανών Στήριξης Διανυσμάτων και Κ - κοντινότερου γείτονα. Η τροποποιημένη μέθοδος τεχνητών αυτοάνοσων συστημάτων (Artificial Immune Systems – AIS) χρησιμοποιήθηκε για πρώτη φορά στην ταξινόμηση και παρουσίασε ιδιαίτερα βελτιωμένα αποτελέσματα στην ταξινόμηση των επιχρισμάτων τα οποία χαρακτηρίζονται ύποπτα (suspicious) από τους ειδικούς και αποτελούν το αδύναμο σημείο της μεθόδου FNA. Αυτές οι περιπτώσεις υπόνοιας αποτελούν ένα πολύ δύσκολο κομμάτι για τη διάκριση μεταξύ των καλοηθειών και των κακοηθειών, ακόμα και για τους πλέον ειδικούς. Επειδή όλα τα περιστατικά που χαρακτηρίζονται από την βιοψία FNA ως υπόνοιες αντιμετωπίζονται κλινικά σαν κακοήθειες, η εφαρμογή των αλγοριθμικών μεθόδων βελτιώνει αισθητά τη διαχείριση αυτών των περιπτώσεων μειώνοντας τον αριθμό των άσκοπων χειρουργικών επεμβάσεων θυρεοειδεκτομών. / The Aim of present thesis is the development of an integrated system for supporting diagnosis (Decision Support System - DSS) using for categorizing FNA biopsy smears. Two categories were selected for the FNA smears: malignant and nonmalignant. The system is constituted by the following stages of 1) data collection, 2) data selection 3) choice of suitable clinical and cytological features, 4) application of data mining method for the categorization of FNA biopsy smears. Furthermore a fundamental objective of the doctoral thesis was the improvement of suspect smears (suspicious) categorization, for the latter FNA Biopsy has a known restriction. The system had been trained and checked in relation to the sample that histologic evaluation existed (ground truth). For smears that characterized as nonmalignant by FNA and histological data we’re not available, complementary clinical, laboratory and imaging evaluations took into account in order to create the sample. Τhe smears that were available in this thesis, were collected from FNA biopsies in Pathologoanatomy Laboratory, A’ Pathology Department, Medical School of Athens University. Given that the above referred laboratory is a reference center, an important number of FNA smears were sent to it from other laboratories for cross check. The examination files were sorted in chronological order, but there were in paper forms. The requirements for the formation and the design of database system were collected. Based on the material of the diagnosis an improved system was designed and developed for data initialization and coding. The database was developed based on the design and analysis of requirements; in this database data were stored for further investigation. Analysis of the graphical user interface design was performed in parallel to the database design. Taking into account that the system might be used after the completion of thesis, the graphical user interface was designed in order to be user friendly and flexible environment. According to the methodological approach that was followed, the various cytological characteristic of 9102 FNA smears aspired among 2000-2004 was analyzed statistically. The cytological reports cross correlated with histological diagnoses, aiming to calculate the effect or contribution of each cytological characteristic to a false or true cytological diagnosis and to find the possible sources of erroneous diagnosis. The smears that have blood or a few follicular cells without colloid were characterized as insufficient for further diagnosis. The aspiration was performed either in Α’ department of Athens University (most of the cases with palpable nodules) or elsewhere (mainly under guidance of the reference center). The acquired smears being send to the reference center from various hospitals with different protocols concerning criteria to perform a thyroid FNA. Histological reports were available for 266 patients. The small number of histological verifications was due to the heterogeneity and the lack of patients files. For evaluating of data, descriptive statistic values were used like mean, standard deviation, percentage, maximum and minimum. In addition to that χ2 tests of significance were performed in order to check possible correlation or independence. For correlating cytological and histological diagnosis and evaluating laboratory findings, apart from the descriptive statistic parameters also calculated sensitivity, specificity, total accuracy, negative predictive value and positive predictive value. Method of statistical significance in the level of 5% (p < 0,05) was applied in order to specify if a disease was correlated to a cytological parameter. Those checks were performed for each disease category in correlation to any cytological parameter. Statistical analysis divided the smears into nonmalignant, malignant, neoplasms, suspicious for malignancy and borderline. A diagnosis support system was implemented using data mining methods. The system is consisted of four stages. The First stage of the system is the Data Collection environment, which stores the data to the database. The Second stage of this system concerns the Selection of Data. User requirements concluded that 111 characteristics are needed to describe each patient (record). Most of them have binary values, presenting existence and not existence, other have alphanumeric and number values. Among them 60 were selected and 7 more are produced from grouping other characteristics. The final analysis reveals that 67 characteristics of the smears are capable for describing the structure of smears in general. The Third stage of system concerns the Selection of Best Characteristics. Due to the high number of attributes (67 per case), it was essential to eliminate the characteristics that are connected linearly or do not bring diagnostics information. The choice of characteristics applied before the classification, having the aim of discovering a subset of characteristics that optimizes the process of classification. The technique of Sequential Float Forward Search (SFFS) was applied. The number of patients that used was 2,036 (1886 non malignancies and 150 malignancies). Among them all malignancies were histologically confirmed. In addition to that 140 no malignancies were histologically confirmed in correlation to evaluation of clinics, laboratorial and medical image actions (ultrasounds etc.). Among 2.036 smears the 25% used for characteristics selection, 37 smears of Malignant and smears of Non Malignant. The Sequential Float Forward Search (SFFS) Technique, choose the best 12 elements that they reveal high performance to FNA data categorization. The Fourth stage is the Application of Classification using Data Mining Methods or in other words data mining method. For this aim a set of reliable, well confirmed but also modern methods applied. In addition to that the system was trained and was checked using the sample with histological verifications (ground truth). The independent application of four reliable methods, Decision Trees, Artificial Neural Network, Support Vector Machine, and k-NN, resulting to comparable outcomes concerning those of FNA. However, further improvement was achieved with the application of Majority (Majority Vote - CMV) using of previous results of three algorithms Artificial Neural Network, Support Vector Machine, and k-NN. The modified Artificial Immune System (AIS) was applied for first time. AIS presents particularly improved results for the categorization of smears, which are characterised “suspicious” by the experts and is a known weakness of FNA method. These cases constitute a very difficult part for the discrimination among non-malignant and malignant, even for a specialist. Since all these cases are faced clinically using FNA as malignancies, the application of an improved algorithmic method improves accordingly the management of these cases by decreasing the number of useless surgical thyroid operations.