L'objectif du travail de thèse, dans le cadre du projet Européen DEMU2NECK, a été de contribuer à développer un modèle volumique déformable et personnalisable du cou, incluant la prise en compte de la capacité d'activation musculaire. Ce modèle vise à contribuer à une meilleure compréhension des liens entre pathologie, posture et activation musculaire pour aider les cliniciens et les fabricants de dispositifs médicaux dans leurs prises de décision. Pour atteindre cet objectif la thèse a été structurée en quatre grandes tâches : après une i) synthèse bibliographique, ii) un modèle générique a été développé et validé. La troisième étape a été iii) le développement d'un modèle de muscle actif et son intégration au modèle générique. L'implémentation de la fonction contractile des muscles lors de simulations musculo-squelettiques isométriques et dynamiques de tâches fonctionnelles simples du cou, a permis de mettre en avant des efforts transverses transmis par les muscles au rachis, allant dans le sens de l'hypothèse d'une contribution à sa stabilisation. La dernière tâche a été iv) la réalisation des premières applications de personnalisation, qui ont porté sur la déformation du modèle générique sur la base de données d'imagerie obtenues sur volontaires. Des comparaisons des réponses des modèles personnalisés et du modèle générique ont mis en avant l'intérêt scientifique de l'approche de personnalisation à travers l'obtention de différences significatives de résultats / The objective of this thesis, as part of the DEMU2NECK European project, was to contribute to develop a 3D, deformable model of the neck, with contractile muscles and including the possibility of patient-specific geometric personalisation. The aim of this model is to contribute to a better understanding of the links that exist between pathologies, cervical spine posture and muscular activation in order to help clinicians and medical devices manufacturers in their decision making process. To achieve this goal, the thesis work was divided into four main tasks: after a i) bibliographic synthesis, ii) a passive generic finite element model was developed and validated. The third step consisted in iii) the development of a finite element muscle model and its integration to the generic passive model. The contractile function of the muscles was implemented during isometric and dynamic simulations of simple functional tasks of the neck. The transverse forces that were transmitted from the muscles to the cervical spine are in good agreement with the hypothesis of a contribution of these forces to the cervical spine stability. The last task covered iv) the personalisation process of the generic model. The responses of subject-specific models based on volunteers were compared with the ones obtained from the generic model. They showed significant differences and, therefore, the scientific relevance of the personalization approach

Ces dernières années, la biologie a subi une profonde mutation, impulsée notamment par les technologies à haut débit et la montée de la génomique personnalisée. L’augmentation massive et constante de l’information biologique qui en résulte offre de nouvelles opportunités pour comprendre la fonction et l’évolution des gènes et génomes à différentes échelles et leurs rôles dans les maladies humaines. Ma thèse s’est articulée autour de la relation entre chercheurs et information biologique, et j’ai contribué à (OrthoInspector) ou créé (Parsec, MyGeneFriends) des systèmes permettant aux chercheurs d’accéder, analyser, visualiser, filtrer et annoter en temps réel l’énorme quantité de données disponibles à l’ère post génomique. MyGeneFriends est un premier pas dans une direction passionnante, faire en sorte que ce ne soient plus les chercheurs qui aillent vers l’information, mais que l’information pertinente aille vers les chercheurs sous une forme adaptée, permettant l’accès personnalisé et efficace aux grandes quantités d’informations, la visualisation deces informations et leur interconnexion en réseaux. / In recent years, biology has undergone a profound evolution, mainly due to high through put technologies and the rise of personal genomics. The resulting constant and massive increase of biological data offers unprecedented opportunities to decipher the function and evolution of genes and genomes at different scales and their roles in human diseases. My thesis addressed the relationship between researchers and biological information, and I contributed to (OrthoInspector) or created (Parsec, MyGeneFriends) systems allowing researchers to access, analyze, visualize, filter and annotate in real time the enormous quantity of data available in the post genomic era. MyGeneFriends is a first step in an exciting new direction: where researchers no longer search forinformation, but instead pertinent information is brought to researchers in a suitable form, allowing personalized and efficient access to large amounts of information, visualization of this information,and their integration in networks.

Social media has changed how political communication is distributed (Häussler, 2021) and has led to party leaders’ using the platforms to create an image of themselves as ordinary individuals (Bărbieru, 2020). Ahead of the 2022 election, social media is expected to play a more prominent role, with it being the source of choice for first-time voters (Internetstiftelsen, 2022). Today voters are also said to identify themselves more with the political candidate than the party (Bjerling, 2012) which makes it relevant to study party leaders’ usage of personalisation on social media. The purpose of this study is to examine how young adults perceive and create meaning around Swedish party leaders' self-construction through personalisation on Instagram by combining semi-structured interviews and reception analysis to be able to understand the phenomenon of party leaders’ personalisation on social media through the interviewees perspective. Drawing on theories of Political Impression Management, Impression Management and Parasocial Relations. By examining how the biggest users of social media, young adults (Internetstiftelsen, 2021; 2022), perceive and create meaning around party leaders' self-construction through personalisation on social media, it can show how the party leaders are perceived and if the interviewees perceive the self-construction to be genuine. Eight interviews with young adults between the ages of 21 to 28 years old have been conducted and the material from the interviews has been coded and thematised in NVivo 12.  The results show that self-construction is perceived to be visible when the party leaders show that they are one of the people. The images were interpreted to be genuine through the party leaders sharing their everyday life, this also indicated if the party leader could impact their audience. The conclusion of the study is that self-construction can be visible through the party leaders' usage of relatable aspects to be perceived as one of the people. In this particular study, it could be divided into three main parts: nature, family and selfies. The interviewees perceive Swedish party leaders’ self-construction through personalisation on Instagram to be genuine if it reflected on the party leaders’ everyday life and their spare time outside of politics. If the image was a selfie it was increasingly more perceived as genuine. Even though the reception material was from the party leaders' self-construction through personalisation, it cannot determine if it represents the party leaders’ personal life, or them outside of politics. With all images being perceived by the interviewees as positive for the party leader. / Sociala medier har förändrat hur politisk kommunikation distribueras (Häussler, 2021) och har lett till att partiledarna använder sig av plattformarna för att kunna skapa en bild av sig själva som vanliga individer (Bărbieru, 2020). Inför valet 2022, förväntas sociala medier spela en mer framträdande roll då det är den källan förstagångsväljare i allra första hand väljer (Internetstiftelsen 2022). Idag sägs även väljare identifiera sig mer med den politiska kandidaten än partiet själv (Bjerling, 2012) vilket gör det relevant att studera partiledares användning av personalisering på sociala medier. Syftet med studien är att undersöka hur unga vuxna uppfattar och skapar mening kring svenska partiledares framställning genom personalisering på Instagram genom att kombinera semistrukturerade intervjuer och en receptionsanalys för att förstå fenomenet av partiledares personalisering på sociala medier från intervjupersonernas perspektiv. Samt utgå från teorier om Political Impression Management, Impression Management och Parasocial Relations. Genom att undersöka hur de största användarna av sociala medier, unga vuxna (Internetstiftelsen, 2021; 2022), uppfattar och skapar mening kring partiledares framställning genom personalisering på sociala medier, kan det visa hur partiledarna uppfattas och om intervjupersonerna uppfattar att framställningen är genuin. Åtta intervjuer med unga vuxna mellan 21 och 28 år har genomförts och materialet från intervjuerna har kodats och tematiseras i NVivo 12.  Resultatet visar att framställning uppfattas vara synligt när partiledarna visar att dem är en av befolkningen. Bilderna uppfattades vara genuina genom att partiledarna delade med sig av sin vardag, detta indikerade också på om partiledaren kunde påverka sin publik. Slutsatsen för studien är att personlig framställning kan vara synlig när partiledarna använder sig av relaterbara aspekter för att uppfattas som en i befolkningen. För denna specifika studie kan de delas in i tre huvudsakliga delar: natur, familj och selfies. Intervjupersonerna uppfattar svenska partiledares framställning genom personalisering på Instagram att vara genuin när det reflekterade partiledarnas vardagliga liv och deras fritid utanför politiken. Om bilden var en selfie ökade uppfattningen att den var genuin. Även om receptions materialet var från partiledarnas personliga framställning så kan det inte fastställas att det representerar partiledarnas privatliv eller hur dem är utanför politiken. Då intervjupersonerna uppfattade alla bilder att vara positiva för partiledarna.

The popularity of mobile period tracker apps, designed to help women track their periods and fertility, has skyrocketed over the past decade. The target audience is people who menstruate, most often women. There are numerous articles on personalisation in mobile apps, but personalisation in mobile period tracker apps has been little studied. Therefore, in this thesis, I analyse the ways of collecting information about users and personalising the user's account in the period tracker apps, as well as whether this personalisation meets the needs of female users. My research question is: How could the personalisation of mobile period tracker apps be improved through design to meet real users' needs? The study builds on the user experience (UX) design process and consists of the following phases: UX research, design, and user testing. In addition, it includes an ideation part. I used post-structural feminist theory from the perspective of Judith Butler's work as a framework. I conducted seven semi-structured interviews with female users who actively use period tracker apps. I studied three chosen period tracker apps using the walkthrough method and conducted a co-design workshop using the themes identified through the thematic analysis of the interviews. The research output is a design solution tested on five participants of the study and then iterated. Through my work, I contribute to studies of reproductive technologies and the field of feminist human-computer interaction (HCI) through suggestions on personalisation.

Cancer treatment personalisation is a major objective in radiation oncology, e.g., in order to move from population-based approaches towards tailored interventions such as dose escalation or de-escalation for specific patients or subgroups. HNSCC is a tumour entity that might particularly benefit from such an approach due to survival rates remaining relatively uniform over time even with multimodal treatment. A personalised approach requires biomarker identification to precisely characterize the tumour phenotype and thus prognosticate patient outcome and response to treatment. Biomarkers can be derived from a wide range of sources: clinical records of patients with semantic features, omics data from tumour biopsies indicating patient gene expression and routinely-taken imaging data from which radiomics features might be extracted. Such data can be then be leveraged alongside machine learning algorithms for modelling of various endpoints. The present thesis was dedicated not only towards establishing prognostic models within locally-advanced HNSCC but also towards a radiogenomic study of the relationship and interplay between image-level information and transcriptome-level information. First, two prognostic signatures for LRC within locally-advanced HNSCC were developed and externally validated through specific feature selection and modelling algorithms suggested through previous work (Leger et al, 2020). As biomarkers should take into account various sources of data, such signatures were built around a highly-prognostic clinical feature: the tumour volume. The selected features supplied statistical or textural information related to tumour heterogeneity and were chosen to provide non-redundant information to the tumour volume. Such an approach is aimed at providing added value to already-validated biomarkers and avoids susceptibilities within radiomics signature development (Welch et al.,2019) such as unspecified dependencies and multi-collinearity within models. One longstanding question within radiomics is the relationship between macroscopic-level image features and underlying biological processes (Aerts et al., 2014). In order to answer this question within the context of locally-advanced HNSCC and CT-based radiomics, surrogacy models were developed for six different gene signatures representing different tumour characteristics such as hypoxia or EMT. All modelling approaches showed low to no relationship between expression of such gene signatures and whole-tumour macroscopic imaging features. This might suggest that while whole-tumour radiomics has shown potential for prognostic endpoint modelling, such features may lack potential to serve as surrogates for local assessments of hypoxia or other microscopic tumour characteristics from a biopsy. While whole-tumour radiomics features may not be able to serve as surrogates for specific, microscopic-level tumour characteristics, they might still be used as surrogates for molecular subtyping within HNSCC as has been done in other tumour entities. Models for the classification of the four HNSCC subtypes (atypical, basal, classical and mesenchymal) were developed. Models differentiating the atypical subtype from the others and the atypical from the mesenchymal subtype were externally validated. As not all four subtypes could be properly classified, this might imply that some subtypes are more similar than others on a CT-level and that the atypical subtype would be the most differentiable subtype. As radiomics features showed low association with individual gene expressions, such omics features might be used to bolster radiomics features for LRC prognosis. However, due to the extremely high dimensionality of omics datasets in comparison to sample size, three methods for expression aggregation were leveraged to find either prognostic regulator genes with VIPER, prognostic coregulated gene modules (WCGNA) or prognostic pathway-level metagenes (GSVA). Enhancement of prognostic performance and stratification for LRC was found with the GSVA approach, through metagenes signifying E2F target activity and hedgehog pathway activity, and one WCGNA network module related to histone regulation. This result not only demonstrates the potential of transcriptomics to bolster radiomics but also of specific omics expression aggregation algorithms to synthesise new prognostic features. Finally, radiomics features were leveraged in order to create and externally validate radiophenotypes for HNSCC, creating five radiophenotypes that showed not only different imaging patterns but also different pathway-level expressions and prognosis for LRC and OS. Of those five radiophenotypes, three were able to be translated into an external validation cohort. However, not all translatable radiophenotypes showed similar stratification between cohorts for all endpoints. This, alongside with different expression levels between cohorts, might signify that one radiophenotype points to different levels of expression of different pathways, thus resulting in significantly different stratification for patients with similar imaging patterns. The field of radiogenomics analysis is, however, still in its infancy and offers a fascinating variety of open research questions for further consideration and study, e.g., promising deep-learning architectures such as transformers combining both image and tabular data (Park et al., 2022), multiomics biomarkers incorporating already-existing mechanistic knowledge (Yoo et al., 2022) or region-based radiomics risk assessments (Leger et al., 2020). While individualised treatment of patients is a daunting enterprise requiring a multidisciplinary approach, in this thesis we went a step closer to this goal for locally-advanced HNSCC:Contents iii List of Figures vii List of Tables ix 1 Introduction 1 2 Theoretical background 5 2.1 Head and neck squamous cell carcinoma 5 2.1.1 Diagnosis and treatment 5 2.1.2 Tumorigenesis 6 2.1.3 Molecular characteristics of HNSCC 7 2.1.4 Biomarkers in HNSCC 7 2.2 Physical principles of computerised tomography 10 2.3 Radiomics data 11 2.3.1 Feature extraction 12 2.3.2 Feature computation 12 2.3.3 Laplacian of Gaussian (LoG) 18 2.3.4 Image perturbations and stability 18 2.4 Transcriptomics data analysis 20 2.4.1 Weighted correlation gene network analysis 21 2.4.2 Virtual inference of protein-activity enriched by regulon 22 2.4.3 Gene Set Enrichment Analysis 24 2.4.4 Gene Set Variation Analysis 25 2.5 Fundamentals of survival analysis 27 2.5.1 Cox proportional hazards model 28 2.5.2 Assessing the proportional hazards assumption 29 2.5.3 Performance of survival models 30 2.6 Machine learning framework 32 2.6.1 Data preprocessing 34 2.6.2 Data resampling 36 2.6.3 Feature selection 36 2.6.4 Hyperparameter optimisation 38 2.6.5 Model building 39 2.6.6 Validation 41 3 Definition and validation of a radiomics signature for LRC prognosis in HNSCC 43 3.1 Motivation 43 3.2 Data and experimental design 44 3.2.1 Patient cohort 44 3.2.2 Clinical data 45 3.2.3 Radiomics features extraction and preprocessing 45 3.2.4 Modelling approach 46 3.3 Results 47 3.3.1 Clinical signature 47 3.3.2 Unwindowed signature 48 3.3.3 Windowed signature 50 3.3.4 Windowing effect comparison 52 3.4 Summary and discussion 52 4 CT radiomics surrogacy of gene signatures 55 4.1 Motivation 55 4.2 Data and experimental methods 55 4.2.1 Patient cohort 56 4.2.2 Transcriptome feature extraction 56 4.2.3 Radiomics features extraction and preprocessing 57 4.2.4 Selected gene signatures and class assignment 57 4.2.5 Modelling approach and metrics 58 4.3 Results 59 4.3.1 Cluster classes 59 4.3.2 Surrogate model performance 59 4.4 Summary and discussion 62 5 Radiomics signatures for HNSCC subtype classification 65 5.1 Motivation 65 5.2 Data and experimental design 66 5.2.1 Patient cohort 66 5.2.2 Radiomics image feature extraction 67 5.2.3 Subtype assignment and characterisation 67 5.2.4 Modelling approach 68 5.3 Results 68 5.3.1 Subtype assignment and characterisation 68 5.3.2 OVA model results 70 5.3.3 OVO model results 72 5.3.4 Multiclass model results 73 5.4 Summary and discussion 74 6 Enhancing radiomics using transcriptome-level information 77 6.1 Motivation 77 6.2 Data and experimental design 78 6.2.1 Patient cohort 78 6.2.2 WCGNA module extraction 79 6.2.3 VIPER regulon extraction 79 6.2.4 GSVA pathway feature creation 79 6.2.5 Radiomics image feature extraction 80 6.2.6 Modelling approach and metrics 80 6.3 Results 81 6.3.1 Radiomics signature 81 6.3.2 WCGNA results 82 6.3.3 VIPER results 85 6.3.4 GSVA results 86 6.4 Summary and discussion 87 7 Radiophenotype discovery through corrected consensus clustering in locally ad-vanced HNSCC 89 7.1 Motivation 89 7.2 Data and experimental design90 7.2.1 Patient cohort 91 7.2.2 Radiomics image feature extraction 91 7.2.3 Reduction and clustering of radiomics features 91 7.2.4 Candidate configuration pruning 92 7.2.5 Radiophenotype characterisation 92 7.2.6 Validation of subtypes 93 7.3 Results 93 7.3.1 M3C clustering results and survival-based pruning 93 7.3.2 Radiophenotype characterisation 95 7.3.3 Radiophenotype validation 100 7.4 Summary and discussion 102 8 Summary 105 9 Zusammenfassung 107 Bibliography 111 Appendix 131 Erklärungen 173 / Die Personalisierung der Krebsbehandlung ist ein wichtiges Ziel in der Radioonkologie, z.B. um von populationsbasierten Ansätzen zu maßgeschneiderten Interventionen wie einer Dosiseskalation oder Deeskalation für bestimmte Patienten oder Patientengruppen überzugehen. HNSCC ist eine Tumorentität, die von einem solchen Ansatz besonders profitieren könnte, da die Überlebensraten auch bei multimodaler Behandlung über der Zeit relativ konstant geblieben sind. Ein personalisierter Ansatz erfordert die Identifizierung von Biomarkern, um den Tumorphänotyp genau zu charakterisieren und somit den Behandlungserfolg vorherzusagen. Biomarker können aus einer Vielzahl von Quellen stammen: klinische Charakteristika von Patienten, Omics-Daten aus Tumorbiopsien, oder routinemäßig aufgenommene Bildgebungsdaten, aus denen Radiomics-Variablen extrahiert werden können. Diese Daten werden dann zusammen mit maschinellen Lernalgorithmen zur Modellierung verschiedener Endpunkte genutzt. Die vorliegende Dissertation widmet sich nicht nur der Etablierung prognostischer Modelle für lokal fortgeschrittene HNSCC, sondern auch einer radiogenomischen Untersuchung der Beziehung zwischen Informationen auf Bildebene und Informationen auf Transkriptomebene. Zunächst wurden zwei prognostische Signaturen für die LRC bei lokal fortgeschrittenem HNSCC entwickelt und extern validiert, unter Verwendung spezifischer Methoden zur Variablenselektion und Modellierungsalgorithmen, die durch frühere Arbeiten von Leger et al. (Leger et al., 2019) vorgeschlagen wurden. Die Signaturen wurden unter Berücksichtigung eines hochgradig prognostischen klinischen Merkmals, dem Tumorvolumen, entwickelt. Die ausgewählten Variablen beschreiben statistische oder strukturelle Informationen in Bezug auf die Tumorheterogenität und wurden so ausgewählt, dass sie zusammen mit dem Tumorvolumen nicht-redundante Informationen liefern. Ein solcher Ansatz zielt darauf ab, bereits validierten Biomarkern einen Mehrwert zu bieten und Anfälligkeiten bei der Entwicklung von Radiomics-Signaturen zu vermeiden (Welch et al., 2019), wie nicht-spezifizierte Abhängigkeiten und Multikollinearität innerhalb von Modellen. Eine langjährige Frage im Feld der Radiomics ist die Beziehung zwischen Bildvariablen auf makroskopischer Ebene und den zugrunde liegenden biologischen Prozessen (Aerts et al., 2014). Um diese Frage für lokal fortgeschrittene HNSCC und CT-basierte Radiomics zu beantworten, wurden Radiomics-basierte Vorhersagemodelle für sechs verschiedene Gensignaturen entwickelt, die unterschiedliche Tumorvariablen wie Hypoxie oder EMT repräsentieren. Alle Modellierungsansätze zeigten eine geringe Korrelation zwischen der Expression solcher Gensignaturen und makroskopischen Variablen der CT-Bildgebung des gesamten Tumors. Dies könnte darauf hindeuten, dass die Radiomics-Parameter zwar Potenzial für die prognostische Outcomemodellierung zeigen, jedoch möglicherweise nicht als Ersatz für die Beurteilungen von Hypoxie oder anderen mikroskopischen Tumoreigenschaften dienen können. Möglicherweise können Radiomics-Parameter jedoch als Ersatz für die molekulare Subtypisierung von HNSCC verwendet werden, wie dies bei anderen Tumorentitäten der Fall ist. Daher wurden Modelle für die Klassifikation der vier HNSCC-Subtypen (atypisch, basal, klassisch und mesenchymal) entwickelt. Erfolgreich validiert werden konnte die Unterscheidung zwischen dem atypischen Subtyp von den anderen sowie zwischen dem atypischen und mesenchymalen Subtyp. Dass jedoch nicht alle vier Subtypen richtig klassifiziert werden konnten, könnte bedeuten, dass einige Subtypen auf CT-Ebene ähnlicher sind als andere und dass der atypische Subtyp der am besten unterscheidbare ist. Da Radiomics-Variablen eine geringe Assoziation mit individuellen Genexpressionen zeigten, könnten solche Omics-Variablen verwendet werden, um Radiomics-Variablen für die LRC-Prognose zu ergänzen. Aufgrund der hohen Dimensionalität der vorliegenden Transkriptom-daten im Vergleich zur Stichprobengröße wurden drei Methoden zur Expressionsaggregation genutzt, um entweder prognostische Regulatorgene mit dem VIPER-Algorithmus, prognostische koregulierte Genmodule mit dem WCGNA-Algorithmus oder prognostische Pathway-Level-Metagene mit dem GSVA-Algorithmus zu finden. Eine Verbesserung der prognostischen Leistung und Stratifizierung für LRC wurde mit dem GSVA-Ansatz anhand von Metagenen, die E2F-Zielaktivität und Hedgehog-Pathway-Aktivität beschreiben, sowie für ein WCGNA-Netzwerkmodul, das sich auf die Histonregulation bezieht, beobachtet. Dieses Ergebnis zeigt nicht nur das Potenzial der Kombination von Transkriptomics und Radiomics, sondern auch von spezifischen Omics-Expressionsaggregationsalgorithmen zur Synthese neuer prognostischer Variablen. In der abschließenden Studie wurden Radiomics-Variablen genutzt, um Radiophänotypen für HNSCC zu erstellen und extern zu validieren. Fünf Radiophänotypen wurden definiert, die nicht nur unterschiedliche Bildgebungsmuster, sondern auch unterschiedliche Genexpressionen auf Pathway-Ebene sowie Unterschiede in LRC und OS zeigten. Von diesen fünf Radiophänotypen konnten drei in eine externe Validierungskohorte überführt werden, die dort jedoch teilweise eine andere Patientenstratifizierung für die betrachteten Endpunkte zeigten. Dies könnte, zusammen mit unterschiedlichen Expressionsniveaus zwischen den Kohorten, bedeuten, dass ein Radiophänotyp mit unterschiedlichen Expressionsniveaus verschiedener Signalwege assoziiert ist, was zu einer signifikant unterschiedlichen Stratifizierung für Patienten mit ähnlichen Bildgebungsmustern führt. Das Gebiet der radiogenomischen Analyse befindet sich noch in den Anfängen und bietet vielfältige offene Forschungsfragen für zukünftige Studien. Beispielsweise können vielversprechende Deep-Learning-Architekturen wie Transformer untersucht werden, die sowohl Bilddaten als auch tabellierte Daten kombinieren (Park et al., 2022). Weiterhin könnte die Multiomics-Biomarker-Entwicklung unter Einbeziehung bereits vorhandener mechanistischer Erkenntnisse untersucht (Yoo et al., 2022) oder die Radiomics-Risikobewertung basierend auf verschiedenen Tumorregionen (Leger et al., 2020) durchgeführt werden. Während die individualisierte Behandlung von Patienten ein langfristiges Ziel darstellt, das einen multidisziplinären Ansatz erfordert, sind wir diesem Ziel mit dieser Arbeit für lokal fortgeschrittene HNSCC einen Schritt näher gekommen.:Contents iii List of Figures vii List of Tables ix 1 Introduction 1 2 Theoretical background 5 2.1 Head and neck squamous cell carcinoma 5 2.1.1 Diagnosis and treatment 5 2.1.2 Tumorigenesis 6 2.1.3 Molecular characteristics of HNSCC 7 2.1.4 Biomarkers in HNSCC 7 2.2 Physical principles of computerised tomography 10 2.3 Radiomics data 11 2.3.1 Feature extraction 12 2.3.2 Feature computation 12 2.3.3 Laplacian of Gaussian (LoG) 18 2.3.4 Image perturbations and stability 18 2.4 Transcriptomics data analysis 20 2.4.1 Weighted correlation gene network analysis 21 2.4.2 Virtual inference of protein-activity enriched by regulon 22 2.4.3 Gene Set Enrichment Analysis 24 2.4.4 Gene Set Variation Analysis 25 2.5 Fundamentals of survival analysis 27 2.5.1 Cox proportional hazards model 28 2.5.2 Assessing the proportional hazards assumption 29 2.5.3 Performance of survival models 30 2.6 Machine learning framework 32 2.6.1 Data preprocessing 34 2.6.2 Data resampling 36 2.6.3 Feature selection 36 2.6.4 Hyperparameter optimisation 38 2.6.5 Model building 39 2.6.6 Validation 41 3 Definition and validation of a radiomics signature for LRC prognosis in HNSCC 43 3.1 Motivation 43 3.2 Data and experimental design 44 3.2.1 Patient cohort 44 3.2.2 Clinical data 45 3.2.3 Radiomics features extraction and preprocessing 45 3.2.4 Modelling approach 46 3.3 Results 47 3.3.1 Clinical signature 47 3.3.2 Unwindowed signature 48 3.3.3 Windowed signature 50 3.3.4 Windowing effect comparison 52 3.4 Summary and discussion 52 4 CT radiomics surrogacy of gene signatures 55 4.1 Motivation 55 4.2 Data and experimental methods 55 4.2.1 Patient cohort 56 4.2.2 Transcriptome feature extraction 56 4.2.3 Radiomics features extraction and preprocessing 57 4.2.4 Selected gene signatures and class assignment 57 4.2.5 Modelling approach and metrics 58 4.3 Results 59 4.3.1 Cluster classes 59 4.3.2 Surrogate model performance 59 4.4 Summary and discussion 62 5 Radiomics signatures for HNSCC subtype classification 65 5.1 Motivation 65 5.2 Data and experimental design 66 5.2.1 Patient cohort 66 5.2.2 Radiomics image feature extraction 67 5.2.3 Subtype assignment and characterisation 67 5.2.4 Modelling approach 68 5.3 Results 68 5.3.1 Subtype assignment and characterisation 68 5.3.2 OVA model results 70 5.3.3 OVO model results 72 5.3.4 Multiclass model results 73 5.4 Summary and discussion 74 6 Enhancing radiomics using transcriptome-level information 77 6.1 Motivation 77 6.2 Data and experimental design 78 6.2.1 Patient cohort 78 6.2.2 WCGNA module extraction 79 6.2.3 VIPER regulon extraction 79 6.2.4 GSVA pathway feature creation 79 6.2.5 Radiomics image feature extraction 80 6.2.6 Modelling approach and metrics 80 6.3 Results 81 6.3.1 Radiomics signature 81 6.3.2 WCGNA results 82 6.3.3 VIPER results 85 6.3.4 GSVA results 86 6.4 Summary and discussion 87 7 Radiophenotype discovery through corrected consensus clustering in locally ad-vanced HNSCC 89 7.1 Motivation 89 7.2 Data and experimental design90 7.2.1 Patient cohort 91 7.2.2 Radiomics image feature extraction 91 7.2.3 Reduction and clustering of radiomics features 91 7.2.4 Candidate configuration pruning 92 7.2.5 Radiophenotype characterisation 92 7.2.6 Validation of subtypes 93 7.3 Results 93 7.3.1 M3C clustering results and survival-based pruning 93 7.3.2 Radiophenotype characterisation 95 7.3.3 Radiophenotype validation 100 7.4 Summary and discussion 102 8 Summary 105 9 Zusammenfassung 107 Bibliography 111 Appendix 131 Erklärungen 173

No description available.

Moderne Fahrzeuginformationssysteme ermöglichen es dem Fahrer, während der Fahrt auf eine Vielzahl an geographischen Informationen wie Sonderziele zuzugreifen. Die Auswahl einer Option aus solch umfangreichen Angeboten während der Fahrt kann jedoch zu einem erheblichen Sicherheitsrisiko werden, welches eine Unterstützung des Fahrers bei der Optionsauswahl erforderlich macht. Eine vielversprechende Möglichkeit der Unterstützung liegt im Anbieten von Assistenzsystemen zur personalisierten Sonderzielauswahl. Ziel der hier vorliegenden Dissertation ist die Optimierung der Akzeptanz solcher Assistenzsysteme im fahrzeugspezifischen Nutzungskontext. Hierzu wurde in einer qualitativen Studie zunächst der Nutzungskontext untersucht. Aufbauend auf den Ergebnissen und der Taxonomie für Assistenzsysteme nach Wandke (2005) wurden drei Assistenzarten mit verschiedenen Assistenzgraden identifiziert, die potentiell für die personalisierte Optionsauswahl im Fahrzeug in Frage kommen (adaptierbare Filterassistenz, adaptive Filterassistenz, adaptive Delegationsassistenz). In zwei Szenariobefragungen konnte mit Hilfe des UbiTAM nach Rothensee (2010) aufgezeigt werden, dass alle drei Assistenzarten positiv bewertet wurden. Innerhalb dieser positiven Bewertungen wurden höhere Assistenzgrade signifikant besser bewertet. Basierend auf den Erkenntnissen wurden die Assistenzarten prototypisch realisiert und in zwei Fahrsimulationsstudien hinsichtlich ihrer Auswirkung auf die Fahrleistung und die Güte der Aufgabenbearbeitung untersucht. In den Studien konnten die Ergebnisse zur Akzeptanzbeurteilung bestätigt werden. Zusätzlich wurde deutlich, dass die adaptiven Assistenzarten das höchste Maß an Unterstützung in der Nebenaufgabe boten. Gleichzeitig zeigten sie auch die geringste Auswirkung auf die Fahrleistung. Die Ergebnisse dieser Arbeit legen für die Gestaltung von Auswahlassistenzsystemen im Fahrzeug nahe, eine Kombination aus adaptiven Filter- und Delegationsassistenzen anzustreben. / Modern vehicle information systems allow the driver to access a variety of geographical information, like points-of-interest, whilst driving. However, selecting an option from an extensive list of options, whilst driving, can pose a significant safety hazard. Thus, it becomes essential to support the driver in this selection process. A promising solution lies in supplying an assistance system that includes a personalised points-of-interest selection. The purpose of this doctoral thesis is to optimise the acceptance of those assistance systems. Firstly, a qualitative study was carried out to examine the context of use. The results were analysed using the taxonomy for assistance systems developed by Wandke (2005). This identified three suitable types of option selection assistance systems, with varying levels of support; adaptable filter assistance, adaptive filter assistance, adaptive delegation assistance. Secondly, the assistance systems were analysed by means of scenario surveys based on the UbiTAM, by Rothensee (2010). The results showed that all three assistance systems were rated positive. Within these positive ratings, those systems offering higher levels of support received significantly better ratings. Based on the findings all three assistance systems were implemented as prototypes. The prototypes were evaluated in two driving simulator studies, in order to assess their effects on both the driving performance and the quality of task processing. The simulator studies verified the results uncovered by the scenario surveys. Additionally, it became clear that adaptive assistance systems provided drivers with the most support during the fulfilment of secondary tasks. Finally, they showed that driving performance was only slightly affected by the use of the adaptive assistance systems. In terms of the development of assistance systems, the results of this thesis suggest that a combination of adaptive filter and delegation assistance should be implemented.

In today’s society there is a discussion regarding students’ level of knowledge and personalisation. The classrooms consist of heterogenic groups of students with different learning needs and levels of knowledge. The purpose of the present study is to find out how the students who participated perceive the teaching of English as an additional language with regards to personalisation, and how they believe differentiation can help develop and deepen their learning.   This study has been conducted using a qualitative method with interviews as the core, complemented by observations. The results have been analysed and interpreted hermeneutically.   The results show that those students who have a positive attitude to English education, irrespective of their grade, are those who see the benefit of knowing English outside the classroom, and thereby create a connection between reality and studying English. Those who lack this connection have a negative attitude. Basing teaching in reality renders personalisation viable for the individual within a teaching group. Even when the students give some examples of how the teaching can be adjusted to their needs, most of them believe that the teacher must adjust the teaching to fit the class as a whole and they do not expect any personalisation. / I dagens samhällsdebatt om skolan diskuteras elevers kunskapsnivå och möjligheten till individanpassning. Klassrummen är sammansatta av heterogena grupper bestående av elever med olika inlärningsbehov och på olika kunskapsnivåer. Syftet med studien är att undersöka hur några betygsmässigt hög- och lågpresterande gymnasieelever upplever att engelskundervisningen är individanpassad efter deras behov, samt hur de anser att den ska vara utformad för att de ska kunna utveckla och fördjupa sitt lärande.   Undersökningen har genomförts utifrån en kvalitativ metod med intervjuer som det centrala och med observationer som bakgrund. Resultatet har analyserats och tolkats hermeneutiskt.   I studien framkom att elever som har en positiv inställning till engelska, oberoende av betyg, är de som ser nyttan utanför klassrummets väggar och därmed verklighetsanknyter engelskundervisningen. De som saknar en koppling till verkligheten har en genomgående negativ inställning. Det är således genom verklighetsanknytningen som individanpassningen möjliggörs. Även om eleverna kommer med förslag på hur undervisningen kan tänkas anpassas till dem, säger flera av dem att läraren måste anpassa undervisningen till klassen som helhet och de förväntar sig ingen individanpassning.

<p>In today’s society there is a discussion regarding students’ level of knowledge and personalisation. The classrooms consist of heterogenic groups of students with different learning needs and levels of knowledge. The purpose of the present study is to find out how the students who participated perceive the teaching of English as an additional language with regards to personalisation, and how they believe differentiation can help develop and deepen their learning.</p><p> </p><p>This study has been conducted using a qualitative method with interviews as the core, complemented by observations. The results have been analysed and interpreted hermeneutically.</p><p> </p><p>The results show that those students who have a positive attitude to English education, irrespective of their grade, are those who see the benefit of knowing English outside the classroom, and thereby create a connection between reality and studying English. Those who lack this connection have a negative attitude. Basing teaching in reality renders personalisation viable for the individual within a teaching group. Even when the students give some examples of how the teaching can be adjusted to their needs, most of them believe that the teacher must adjust the teaching to fit the class as a whole and they do not expect any personalisation.</p><p> </p> / <p>I dagens samhällsdebatt om skolan diskuteras elevers kunskapsnivå och möjligheten till individanpassning. Klassrummen är sammansatta av heterogena grupper bestående av elever med olika inlärningsbehov och på olika kunskapsnivåer. Syftet med studien är att undersöka hur några betygsmässigt hög- och lågpresterande gymnasieelever upplever att engelskundervisningen är individanpassad efter deras behov, samt hur de anser att den ska vara utformad för att de ska kunna utveckla och fördjupa sitt lärande.</p><p> </p><p>Undersökningen har genomförts utifrån en kvalitativ metod med intervjuer som det centrala och med observationer som bakgrund. Resultatet har analyserats och tolkats hermeneutiskt.</p><p> </p><p>I studien framkom att elever som har en positiv inställning till engelska, oberoende av betyg, är de som ser nyttan utanför klassrummets väggar och därmed verklighetsanknyter engelskundervisningen. De som saknar en koppling till verkligheten har en genomgående negativ inställning. Det är således genom verklighetsanknytningen som individanpassningen möjliggörs. Även om eleverna kommer med förslag på hur undervisningen kan tänkas anpassas till dem, säger flera av dem att läraren måste anpassa undervisningen till klassen som helhet och de förväntar sig ingen individanpassning.</p>

Personalizuotos mokymosi aplinkos kūrimas parenkant besimokančiajam tinkamas internetines priemones yra sudėtingas ir aktualus šių dienų uždavinys. Dabartinis besimokantysis turėtų pats imtis iniciatyvos, būti atsakingas už mokymosi procesą, mokėti pasirinkti tinkamas mokymosi priemones, tačiau parama mokymosi metu irgi yra labai svarbi, nes abejotina, ar besimokančiojo savarankiškai pasirinktos priemonės padės įgyvendinti siekiamų mokymosi tikslų ir veiklų optimaliausiu būdu. Disertaciniame darbe išnagrinėti rekomendavimo sistemų ir jose naudojamų vartotojo profilių tipai, rekomendavimo būdai, šių sistemų taikymo galimybės personalizuotam mokymuisi, mokymosi proceso ir antrosios kartos saityno priemonių sąveika, pagrindinės savybės. Pateikiamas antrosios kartos saityno priemonių komponavimo mokymosi procese metodas parenkantis tam tikram besimokančiajam priemonę atsižvelgus į mokymosi tikslus, norimą įgyvendinti mokymosi veiklą, teikiamą pirmenybę mokymosi turiniui bei bendravimo formai. Nagrinėtoji dalykinė sritis aprašyta ontologijoje, o pasiūlyto metodo etapai įgyvendinti žiniomis grindžiamos rekomendavimo sistemos prototipe. Sukurtoji sistema rekomenduoja tam tikrą mokymosi stilių turinčiam besimokančiajam visas jos žinių bazėje esančias internetines priemones, kuriomis naudodamasis besimokantysis gali atlikti nurodytą mokymosi veiklą. Pasiūlytą metodą įvertino parinkti ekspertai – buvo įsitikinta metodo kokybe, t. y. tikslumu, tinkamumu ir našumu (laiko atžvilgiu)... [toliau žr. visą tekstą] / Although in practice and in many sources of literature the need for the application of Web 2.0 tools in education is highlighted, however there is lack of clear methods how these tools could be applied in learning for a higher learning quality, and there is a lack of e-learning systems that implement personalised Web 2.0 tools selection. In the work, personalised Web 2.0 tools selection method is presented. In the research, first of all, personalised e-learning technological peculiarities i.e. recommender systems applications for learning personalisation and those systems components were investigated. After that, selection methods for Web 2.0 tools suitable for implementing learning activities were analysed. The method of integrating Web 2.0 tools into personalised learning activities according to students learning styles was created (this method takes into account student’s learning preferences for content and communication modes tailored to the learning activities with a view to help the learner to quickly and accurately find the right educational tools) and prototype of the recommender system that implements the method proposed was developed. Finally, the expert evaluation of the developed system prototype that implements the method proposed was performed.