Un cadre statistique bayésien a été créé dans le cadre de cette thèse pour le développement de modèles cliniques basés sur une approche d’apprentissage continu dans laquelle de nouvelles données peuvent être ajoutées. L’objectif des modèles est de prévoir les effets de la radiothérapie à partir de preuves cliniques. Des concepts d’apprentissage machine ont été utilisés pour résoudre le cadre bayésien. Les modèles développés concernent un cancer du cerveau agressif appelé glioblastome. Les données médicales comprennent une base de données d’environ 90 patients souffrant de glioblastome ; la base de données contient des images médicales et des entrées de données telles que l’âge, le sexe, etc. Des modèles de prévision neurologique ont été construits pour illustrer le type de modèles qui sont obtenus avec la méthodologie. Des modèles de récidive du glioblastome, sous la forme de modèles linéaires généralisés (GLM) et de modèles d’arbres de décision, ont été développés pour explorer la possibilité de prédire l’emplacement de la récidive à l’aide de l’imagerie préradiothérapie. Faute d’une prédiction suffisamment forte obtenue par les modèles arborescents, nous avons décidé de développer des outils de représentation visuelle. Ces outils permettent d’observer directement les valeurs d’intensité des images médicales concernant les lieux de récidive et de non-récurrence. Dans l’ensemble, le cadre élaboré pour la modélisation des données cliniques en radiothérapie fournit une base solide pour l’élaboration de modèles plus complexes. / A Bayesian statistics framework was created in this thesis work for developing clinical based models in a continuous learning approach in which new data can be added. The objective of the models is to forecast radiation therapy effects based on clinical evidence. Machine learning concepts were used for solving the Bayesian framework. The models developed concern an aggressive brain cancer called glioblastoma. The medical data comprises a database of about 90 patients suffering glioblastoma; the database contains medical images and data entries such as age, gender, etc. Neurologic grade predictions models were constructed for illustrating the type of models that can be build with the methodology. Glioblastoma recurrence models, in the form of Generalized Linear Models (GLM) and decision tree models, were developed to explore the possibility of predicting the recurrence location using pre-radiation treatment imaging. Following, due to the lack of a sufficiently strong prediction obtained by the tree models, we decided to develop visual representation tools to directly observe the medical image intensity values concerning the recurrence and non-recurrence locations. Overall, the framework developed for modeling of radiation therapy clinical data provides a solid foundation for more complex models to be developed.

Glioblastoma is the most common and lethal malignant brain tumor with a survival rate of 14.6 months and a tumor recurrence rate of ninety percent. Two key causes for glioblastomas grim outcome derive from the lack of applicable prognostic markers and effective therapeutic targets. By employing a loss of function RNAi screen in glioblastoma cells we found a list of 20 kinases that can be considered glioblastoma survival kinases. These survival kinases which we term as survival kinase genes, (SKGs) were investigated to find prognostic markers as well as therapeutic targets for glioblastoma. Analyzing these survival kinases in The Cancer Genome Atlas patient database, we found that CDCP1, CDKL5, CSNK1𝜀, IRAK3, LATS2, PRKAA1, STK3, TBRG4, and ULK4 genes could be used as prognostic markers for glioblastoma with or without temozolomide chemotherapeutic treatment as a covariate. For the first time, we found that patients with increased levels of NEK9 and PIK3CB mRNA expression had a higher probability of recurrent tumors. We also discovered that expression of CDCP1, IGF2R, IRAK3, LATS2, PIK3CB, ULK4, or VRK1 in primary glioblastoma tumors was associated with tumor recurrence prognosis. To note, of these recurrent prognostic candidates, PIK3CB expression in recurrent tumor tissue had much higher expression compared to primary tissue. Further investigation in the PI3K pathway showed a strong correlation with recurrence rate, days to recurrence and survival emphasizing the role of PIK3CB in tumor recurrence in glioblastoma. In efforts to find effective therapeutic targets for glioblastoma we used SKGs as potential candidates. We chose the serine/threonine kinase, Casein Kinase 1 Epsilon (CSNK1𝜀) as a target for glioblastoma because multiple shRNAs targeted this gene in our loss of function screen and multiple commercially available inhibitors of this gene are available. Casein kinase 1 epsilon protein and mRNA expression were investigated using computational tools. It was revealed that CSNK1𝜀 expression has higher expression in glioblastoma than normal tissue. To further examine this gene we knocked down (KD) or inhibited CSNK1𝜀 in glioblastoma cells lines and noticed a significant increase in cell death without any significant effect on normal cell lines. KD and inhibition of CSNK1𝜀 in cancer stem cells, a culprit of tumor recurrence, also revealed limited self-renewal and proliferation in cancer stem cells and a significant decrease in cell survival without affecting normal stem cells. Further analysis of downstream effects of CSNK1𝜀 knockdown and inhibition indicate a significant increase in the protein expression of β-catenin (CTNNB1). We found that CSNK1𝜀 KD activated β-catenin, which increased GBM cell death, but can be rescued using CTNNB1 shRNA. Our survival kinase screen, computational analyses, patient database analyses and experimental methods contributed to the discovery of novel prognostic markers and therapeutic targets for glioblastoma. / Ph. D.

Orientadores: Ilka de Fátima Santna Ferreira Boin, Cecilia Amélia Fazzio Escanhoela / Tese (doutorado) - Universidade Estadual de Campinas, Faculdade de Ciências Médicas / Made available in DSpace on 2018-08-21T02:58:30Z (GMT). No. of bitstreams: 1 Ataide_ElaineCristinade_D.pdf: 2429864 bytes, checksum: ec486a6adc404d0751bb752ea469797b (MD5) Previous issue date: 2012 / Resumo: Introdução: Apesar de sobrevida ao redor de 70% em cinco anos a recidiva do CHC vem suscitando cuidados com índices que variam na literatura entre 6% a 26%. Muitos são os fatores atrelados ao maior risco de recidiva descritos na literatura, sem definição de qual o melhor método que poderia predizer esse evento de alta letalidade. Objetivo: Os objetivos desse estudo foram: avaliar a sobrevida e recidiva tumoral em pacientes submetidos a transplante hepático por CHC e investigar a imunoexpressão dos marcadores imunohistoquimicos: HSP70, Glipican3, Glutamina sintetase, beta-catenina, CK7, Ck19, HepPar1 e PCNA, estudando sua associação com características tumorais e prognóstico de pacientes submetidos a transplante hepático por CHC. Método: Foram estudados 90 pacientes portadores de CHC submetidos a transplante hepático de 1996 a 2010. Foram estudados fatores correlacionados ao aparecimento da recidiva neoplásica como: tamanho da maior lesão, número de lesões, grau histológico, presença de invasão vascular, nível de alfa-feto proteína (AFP) superior a 200 ng/ml e regime de imunossupressão. Foi estudada também a correlação da expressão dos marcadores imunohistoquimicos estudados com cada uma dessas mesmas variáveis. A técnica de estudo imunohistoquimico foi o arranjo em matriz tecidual (TMA). A análise estatística utilizou testes de regressão uni e multivariadas, teste de Cox, teste de Qui-quadrado ou Fisher, teste de Mann-Whitney e para estudo de sobrevida foi utilizado o método de Kaplan Meyer. Resultados: Foi observada recidiva em 7 pacientes (8%).O tempo de cirurgia, quantidade de concentrados de hemácias administrados, valor do MELD calculado no momento da cirurgia e a presença de recidiva foram associados à menor sobrevida. Pacientes com recidiva tumoral apresentaram tendência à presença de invasão vascular, apresentaram maior número de nódulos e maiores nódulos. Em relação aos marcadores imunohistoquimicos pacientes com glutamina sintetase positiva apresentaram tendência à menor sobrevida; e a presença de HepaPar1 negativo apresentou correlação com o aparecimento de recidiva neoplásica. Pacientes com HSP70 positivo apresentaram maior prevalência do grau histológico III. Pacientes com Glipican3 positivo apresentaram nódulos maiores e presença de mais casos com AFP superior a 200 ng/ml. Pacientes com PCNA positivo apresentaram nódulos maiores. Pacientes com HepPar1 negativo apresentaram nódulos maiores e tendência a apresentar mais nódulos. Pacientes com beta-catenina positiva apresentaram maiores nódulos e presença de maior número de pacientes com grau histológico III. Pacientes com CK19 positivo demonstraram tendência a apresentar nódulos maiores (p=0.05). A associação entre beta-catenina e Glipican3 positivos demonstrou correlação com a presença de nódulos maiores com maior evidência estatística do que quando avaliados separadamente (p=0,003). Conclusão: Não foi possível a associação de nenhum desses marcadores com a sobrevida exceto pela presença de glutamina sintetase positiva que apresentou tendência à associação com piora da sobrevida. A imunoexpressão desses marcadores não se correlacionou com o tempo de aparecimento de recorrência tumoral, com exceção da do HepPar1, o qual, quando negativo, correlacionou-se com maior frequência de aparecimento de recidiva. Entretanto, a maioria dos marcadores estudados apresentaram correlação com pelo menos uma das variáveis em estudo, confirmando nossa hipótese de que esses marcadores podem, sim, auxiliar na avaliação do prognóstico de pacientes submetidos a transplante hepático por CHC / Abstract: Introduction: Although overall survival rates have been around 70% after five years, recurrence of HCC has indices in literature ranging from 6 to 26%. There is no consensus therapy for treatment of HCC recurrence after liver transplantation that could lead to a significant increase in survival. Many factors are linked to higher risk of recurrence in the literature, without defining the best method that could predict this highly lethal event. Objective: The aim of this study was first to evaluate the survival and tumor recurrence in patients undergoing liver transplantation for HCC and second to evaluate immunoexpression of immunohistochemical markers: HSP70, Glipican3, Glutamin synthetase, Beta-Catenin, CK7, Ck19, HepPar1 and PCNA, which are capable of assessing the malignant cellular potential, studying its association with tumor characteristics and prognosis of patients undergoing liver transplantation for HCC. Methods: We studied 90 patients who underwent liver transplantation from 1996 to 2010. We evaluated factors related to survival and tumor recurrence. Factors were also studied related to the onset of neoplastic recurrence as size of the largest lesion, number of lesions, histological grade, presence of vascular invasion, AFP level greater than 200 ng / ml and the immunosuppressive regimen. The correlation of expression of immunohistochemical markers studied was correlated with each of these variables. The immunohistochemistry technique was the tissue matrix arrangement (TMA) and the statistical analysis used was regression testing univariate and multivariate, Cox test, chi-square or Fisher test and Mann-Whitney test, while for study of survival the Kaplan Meyer curve was used. Results: The duration of surgery and recurrence was associated with shorter survival. Patients with tumor recurrence tended to have the presence of vascular invasion, showing a higher number of nodules and larger nodules. Regarding the presence of immunohistochemical markers glutamin synthetase showed a positive trend toward lower survival, and presence of HepPar1 negative correlated with the appearance of neoplastic recurrence. HSP70 positive patients had higher prevalence of histologic grade III. Patients with positive Glipican3 showed larger lesions and more patients had AFP greater than 200 ng / ml. PCNA positive patients had bigger lesions. HepPar1 negative patients had larger lesions and tended to have more nodules. Patients with positive Beta-Catenin showed larger nodules and more histologic grade III. Patients with positive CK19 showed a tendency to have larger nodules. The association between Beta-catenin and Glipican3 showed positive association with larger nodules. Conclusion: There was no statistical correlation of these markers and the specific disease survival except for the presence of glutamine synthetase which showed only a positive trend of association with survival. The immunoreactivity of these markers did not correlate with the time of appearance of a recurrent tumor, with the exception of the Hep-Par1, which, if negative, was correlated with higher frequency of occurrence of relapse. However, most of the markers studied showed correlation with at least one of the variables studied, whether of characteristics of the population (AFP level, the presence of recurrence and survival time) or characteristics of the lesion (tumor number, the greater lesion size, presence of vascular invasion and degree of differentiation) confirming our hypothesis that these markers can indeed assist in assessing the prognosis of patients undergoing liver transplantation for HCC / Doutorado / Fisiopatologia Cirúrgica / Doutor em Ciências da Cirurgia

Purpose: Meningiomas are common intracranial tumors, often well controlled with surgical resection alone. While the efficacy of radiation therapy in improving local control and progression-free survival is well documented, prognostic data substantiate factors that are predictive of poor local control following definitive radiation therapy. PCNA is a DNA polymerase expressed at the highest levels in the S-phase, the most resistant portion of the cell cycle to ionizing radiation in vitro. We investigated the possible correlation between the levels of PCNA expression and the clinical outcome of patients treated with definitive radiation therapy. Methods and Materials: Archival tissue was collected from 33 cases of meningioma treated at our institution for definitive radiation therapy between 1970 and 1990. Age-matched normal meningeal tissue and asymptomatic meningiomas removed at autopsy served as tissue controls. A standard ABC immumoperoxidase technique employing antibodies to PCNA, PC-10 (Dako, California) was used to stain specimen slides for PCNA. PCNA index was defined as the number of positive nuclei per 10 high-power fields at 400x magnification. Two independent observers scored the slides without prior knowledge of the cases at hand. Results: Patients with high PCNA index were less likely to be controlled by therapeutic radiation (p < 0.001, Kaplan- Meier). All patients with a PCNA index greater that 25 failed radiation therapy. Using multivariate analyses, malignant (but not atypical), histology and PCNA index were significant predictors of progression following radiation therapy (p < 0.05, log rank). Conclusion: PCNA index may be a useful adjunct to more standard histopathologic criteria in the determination of meningioma local control and progression-free survival following therapeutic irradiation. Data on a more expanded population evaluated on a prospective basis will be needed before such criteria are routinely employed in the clinical setting.

Cancer remains the second leading cause of death in the United States and the world, despite years of research and the development of different treatments. One reason for this is cancer cells are able to survive through adaptation to their environment and aberrantly activated growth signaling. As such, developing new therapies that overcome these hurdles are necessary to combat cancer. Previous work in our laboratory using RNA interference screening identified genes that regulate the survival of glioblastoma (GBM) or autophagy in chronic myelogenous leukemia (CML) cancer cells. One screen identified Phosphatidylinositol-4,5-bisophosphate 3-kinase catalytic subunit beta (PIK3CB) in the family of Phosphatidylinositol 3-kinases (PI3K) as a survival kinase gene in GBM. Work contained in this dissertation set out to study PIK3CB mediated GBM cell survival. We report that only PIK3CB, in its family of other PI3K genes, is a biomarker for GBM recurrence and is selectively important for GBM cell survival. Another screen identified the long non-coding RNA, Linc00467, as a gene that regulates autophagy in CML. Autophagy is a dynamic survival process used by all cells, benign and cancerous, where cellular components are broken down and re-assimilated to sustain survival. Work contained in this dissertation set out to characterize the role that Linc00467 serves in regulating autophagy in a myriad of cancers. Collectively our data have showed Linc00467 to actively repress levels of autophagy in cancer cells. Further, our data revealed an important role for Linc00467 in regulating the stability of the autophagy regulating protein serine-threonine kinase 11 (STK11). Because of the unique role that Linc00467 serves in regulating autophagy we renamed it as, autophagy regulating long intergenic noncoding RNA or ARLINC. Taken together the work in this dissertation unveils the inner-workings of two important cancer cell survival pathways and shows their potential for development into therapeutic targets to treat cancer. / Ph. D. / Cancer remains the second leading cause of death in the United States and the world, despite years of research and the development of different treatments. One reason for this is cancer cells are able to survive through adaptation to their environment and aberrantly activated growth signaling. As such, developing new therapies that overcome these hurdles are necessary to combat cancer. Previous work in our laboratory using high throughput genetic screens identified genes that regulate the survival of cancer cells from a deadly type of brain cancer called glioblastoma (GBM). Another screen revealed genes that regulate a process called autophagy in cancer cells from a type of leukemia called chronic myelogenous leukemia (CML). Autophagy is a process that cancer cells can use to survive through chemotherapy. One screen identified the gene Phosphatidylinositol-4,5-bisophosphate 3-kinase catalytic subunit β (PIK3CB) in the family of Phosphatidylinositol 3-kinases (PI3K) as a survival gene in GBM. Work contained in this dissertation set out to study PIK3CB mediated GBM cell survival. We report that only PIK3CB, in its family of other PI3K genes, is a biomarker for GBM recurrence and is selectively important for GBM cell survival. Another screen identified the long non-coding RNA, Linc00467, as a gene that regulates autophagy in CML. Autophagy is a dynamic survival process used by all cells, normal and cancerous, where cellular components are broken down and reassimilated to sustain survival. Work contained in this dissertation set out to characterize the role that Linc00467 serves in regulating autophagy in different types of cancer. Collectively our data have showed Linc00467 to actively repress levels of autophagy in cancer cells. Further, our data revealed an important role for Linc00467 in regulating the stability of the autophagy regulating protein serine-threonine kinase 11 (STK11). Because of the unique role that Linc00467 serves in regulating autophagy we renamed it as, autophagy regulating long intergenic noncoding RNA or ARLINC. Taken together the work in this dissertation unveils the inner-workings of two important cancer cell survival pathways and shows their potential for development into therapeutic targets to treat cancer.

Background Current studies indicate that fluorine-18-fluorodeoxyglucose positron emission tomography/ computed tomography ([18F]FDG PET/CT) is the most accurate imaging modality for the detection of relapsed locally advanced non-small cell lung cancer (NSCLC) after curatively intended chemoradiotherapy. To this day, there is no objective and reproducible definition for the diagnosis of disease recurrence in PET/CT, the reading of which is relevantly influenced by post radiation inflammatory processes. The aim of this study was to evaluate and compare visual and threshold-based semi-automated evaluation criteria for the assessment of suspected tumor recurrence in a well-defined study population investigated during the randomized clinical PET-Plan trial. Methods This retrospective analysis comprises 114 PET/CT data sets of 82 patients from the PET-Plan multi-center study cohort who underwent [18F]FDG PET/CT imaging at different timepoints for relapse, as suspected by CT. Scans were first analyzed visually by four blinded readers using a binary scoring system for each possible localization and the associated reader certainty of the evaluation. Visual evaluations were conducted repeatedly without and with additional knowledge of the initial staging PET and radiotherapy delineation volumes. In a second step, uptake was measured quantitatively using maximum standardized uptake value (SUVmax), peak standardized uptake value corrected for lean body mass (SULpeak), and a liver threshold-based quantitative assessment model. Resulting sensitivity and specificity for relapse detection were compared to the findings in the visual assessment. The gold standard of recurrence was independently defined by prospective study routine including external reviewers using CT, PET, biopsies and clinical course of the disease. Results Overall interobserver agreement (IOA) of the visual assessment was moderate with a high difference between secure (ĸ = 0.66) and insecure (ĸ = 0.24) evaluations. Additional knowledge of the initial staging PET and radiotherapy delineation volumes improved the sensitivity (0.85 vs 0.92) but did not show significant impact on the specificity (0.86 vs 0.89). PET parameters SUVmax and SULpeak showed lower accuracy compared to the visual assessment, whereas threshold-based reading showed similar sensitivity (0.86) and higher specificity (0.97). Conclusion Visual assessment especially if associated with high reader certainty shows very high interobserver agreement and high accuracy that can be further increased by baseline PET/CT information. The implementation of a patient individual liver threshold value definition, similar to the threshold definition in PERCIST, offers a more standardized method matching the accuracy of experienced readers albeit not providing further improvement of accuracy.

Objective: Primary transitional cell carcinoma (TCC) of the upper urinary tract represents 6–8% of all TCC cases. Nephroureterectomy with removal of a bladder cuff is the treatment of choice. The rates of TCC recurrence in the bladder after primary upper urinary tract surgery described in the literature range between 12.5 and 37.5%. In a retrospective analysis we examined the occurrence of TCC after nephroureterectomy for upper tract TCC in patients without a previous history of bladder TCC at the time of surgery. Methods: Between 1990 and 2002, 29 patients underwent primary nephroureterectomy for upper tract TCC. The mean age of the patients was 69.5 years. In 5 cases upper urinary tract tumors were multilocular, in the remaining cases unilocular in the renal pelvis (n = 12) or the ureter (n = 12). The follow-up was available for 29 patients with a mean follow-up of 3.37 (0.1–11.2) years. Results: 11/29 (37.9%) patients had TCC recurrence with 9/11 patients having bladder TCC diagnosed within 2.5 years (0.9–6.0) after nephroureterectomy. 13/29 patients are alive without TCC recurrence, 3/29 patients died due to systemic TCC progression and 5/29 died of unrelated causes without evidence of TCC recurrence. Conclusion: Our data indicate a high incidence of bladder TCC after nephroureterectomy for primary upper tract TCC of up to 6 years after primary surgery. Because of the high incidence of bladder TCC within the first 3 years of surgery, careful follow-up is needed over at least this period. / Dieser Beitrag ist mit Zustimmung des Rechteinhabers aufgrund einer (DFG-geförderten) Allianz- bzw. Nationallizenz frei zugänglich.

No / Objectives Previous studies have demonstrated that intravesical administration of apaziquone (EOquin) has ablative activity against superficial bladder cancer marker lesions with 8 out of 12 complete responses recorded. We present a comparison between the rates of tumor recurrence before and after treatment with apaziquone. Methods The rate of tumor recurrence after treatment with apaziquone was compared with each patient's historical record of recurrences obtained from a retrospective analysis of the patients' case notes. The time to each recurrence event before apaziquone treatment and the time to the first recurrence after apaziquone treatment were recorded, and the data were analyzed using a population-averaged linear regression model using Stata Release, version 9.2, software. Results Of the eight complete responses obtained in the Phase I study, tumor recurrence occurred in 4 patients and the remaining 4 patients remained disease free after a median follow-up of 31 months. The time to the first recurrence after apaziquone treatment was significantly longer (P <0.001) compared with the historical pattern and recurrence interval before apaziquone. Before apaziquone instillation, the mean ± SE recurrence rate and tumor rate per year was 1.5 ± 0.2 and 4.8 ± 1.2, respectively, and these decreased to 0.6 ± 0.25 and 1.5 ± 0.8, respectively, after apaziquone treatment (P <0.05). Conclusions The results of this study indicate that early recurrences after treatment with apaziquone are infrequent and the interval to recurrence is significantly greater compared with the historical recurrence times for these patients. Larger prospective randomised trials are warranted to confirm these results. Aapaziquone (EOquin, USAN, E09, 3-hydroxy-5-aziridinyl-1-methyl-2[indole-4,7-dione]¿prop-¿-en-¿-ol) belongs to a class of anticancer agents known as bioreductive drugs that require metabolism by cellular reductases to generate a cytotoxic species.1 Although it is chemically related to mitomycin C, apaziquone has a distinctly different mechanism of action and preclinical activity profile.1 and 2 The initial optimism generated by its preclinical activity profile rapidly evaporated after the demonstration that intravenously administered apaziquone was clinically inactive against a range of solid tumors in Phase II clinical trials.3 and 4 Several possible explanations were considered for its lack of efficacy, but poor drug delivery to the tumor because of the rapid pharmacokinetic elimination of apaziquone in conjunction with relatively poor penetration through avascular tissue was considered to be the principal reason.5 On the basis of the rationale that intravesical administration would circumvent the problem of drug delivery and any apaziquone absorbed into the blood stream would be rapidly cleared,6 a Phase I-II clinical pilot study of intravesical administration of apaziquone to superficial bladder tumors was established.7 The results of that trial demonstrated that intravesically administered apaziquone has ablative activity against superficial bladder transitional cell carcinoma (TCC) marker lesions.7 These results were confirmed and extended in a Phase II clinical trial of 47 patients with superficial bladder TCC, in which complete responses were obtained in 67% of patients.8 Because all the enrolled patients in the original trial7 had had multiple recurrences after previous intravesical chemotherapy and/or immunotherapy, the purpose of the present study was, first, to report the recurrences that occurred after apaziquone treatment and, second, to study the effect of apaziquone instillation on the recurrence rate by statistically comparing these results with the historical pattern of recurrences for each patient before treatment with apaziquone.

O presente trabalho propõe o uso de parâmetros de textura extraídos computacionalmente de IRM como biomarcadores de imagem na predição de recorrência tumoral em pacientes de adenomas pituitários clinicamente não funcionantes (APNF). Para isso, esse estudo analisou imagens de RM de 15 pacientes de APNF retrospectivamente separados em dois grupos: O grupo de pacientes recorrentes, definido por 7 pacientes que exibiram recorrência tumoral em um período de 4, 640 +- 0, 653 anos (média +- erro padrão) de acompanhamento clínico após a primeira abordagem cirúrgica; e o grupo de pacientes estáveis, formado por 8 pacientes com lesões consideradas estáveis em um período de 4,512 +- 0, 536 anos. Uma máscara de segmentação tridimensional da lesão tumoral foi construída manualmente por um especialista sobre a imagem 3D T1-W DCE pré-operatória para cada paciente. Em seguida, essa segmentação e a própria imagem de ressonância foram usadas para extrair 48 características numéricas de textura. Adicionalmente, 4 características clínicas foram consideradas no estudo: a imunohistoquímica, invasividade, idade na primeira cirurgia e sexo, totalizando 52 características. Cada uma destas 52 características fora testada através de testes estatísticos convencionais univariados para ver se existia evidencias do poder discriminatório dessas características para diferenciar esses dois grupos de pacientes. Mais adiante, diferentes subconjuntos dessas características foram usados para construir modelos de predição baseados na teoria de aprendizagem de máquinas (usando os algoritmos k-nearest neighbor (kNN), decision tree (DTC), e random forest (RFC)) para investigar um modelo de classificação capaz de identificar os pacientes que experimentariam recorrência tumoral após a primeira cirurgia. 9 características de textura foram consideradas individualmente significantes (p < 0, 05) na diferenciação dos grupos de paciente recorrente e estável. Afirmando esses achados, a análise com a curva ROC para cada uma das 9 características exibiu medidas de AUC de 0,803 a 0,857 significando uma boa performance de classificação. A idade, imunoistoquímica, invasividade e sexo não mostraram evidencias de associação com recorrência tumoral. As melhores performances com algoritmos de classificação foram com kNN e RFC, ambos atingiram uma especificidade de 1,000 conservando alta acurácia (0,933) e obtendo 0,991 na análise com a curva ROC, o que caracteriza uma performance de classificação quase perfeita. DTC não mostrou nenhuma melhora se comparado com os resultados das classificações univariadas. Esses resultados permitem concluir que parâmetros de textura são úteis na predição de recorrência tumoral após a primeira cirurgia em pacientes de APNF. E que os valores de predição dessas características podem ser observados por testes estatísticos univariados convencionais e por análises multivariadas através de algoritmos baseados em aprendizagem de máquinas / The present work proposes the usage of texture features computationally extracted from MRI as imaging biomarkers in the prediction of tumor recurrence in patients with non-functioning pituitary adenomas (NFPA). With this purpose, this study analyzed MR images from 15 patients of NFPA retrospectively separated in two groups: the recurrent patient group, formed by seven (7) patients who exhibited tumor recurrence in a period of 4,640 +- 0,653 years (mean +- standard error) of follow-up period after the first surgical approach; and the stable patient group formed by eight (8) patients with lesions considered stable in a period of 4,512 +- 0,536 years. A three-dimensional segmentation mask of the tumor lesion was manually performed by a specialist over preoperative 3D T1-W DCE MR image for each patient. Next, this segmentation and the preoperative MRI itself were used to extract 48 numerical textural features. Additionally, 4 clinical features were considered in the study: immunohistochemistry, invasiveness, age at first surgery, and gender, totalizing 52 features. Each one of those 52 features were tested through conventional univariate statistical tests to see if there were evidence of their discrimination power to differentiate these two patient groups. Moreover, different subsets of those features were used to build machine learning prediction models (using k-nearest neighbor (kNN), decision tree (DTC), and random forest (RFC) algorithms) to investigate a classification model capable of identifying the patients that would experience tumor recurrence after the first surgery. 9 quantitative textural features were found to be individually significant (p < 0,05) in the differentiation of recurrent and stable patient group. Affirming these findings, the ROC curve analysis for each one of those 9 features exhibited an AUC score from 0.803 to 0.857 meaning a good classification performance. Age, immunohistochemistry and invasiveness status, and gender did not show evidence of association with tumor recurrence. The best performances with classification algorithms were obtained with kNN and RFC, both reached specificity of 1.000 conserving high accuracy (0.933) and scoring 0.991 in ROC curve analysis, what characterizes an almost perfect classification performance. DTC did not show any improvement compared to the univariate classification results. These findings allow to conclude that textural features are useful in the prediction of tumor recurrence after first surgery in NFPA patients. And that the prediction value of those features can be observed with both conventional univariate statistical tests and multivariate analyses through machine learning algorithms

O presente trabalho propõe o uso de parâmetros de textura extraídos computacionalmente de IRM como biomarcadores de imagem na predição de recorrência tumoral em pacientes de adenomas pituitários clinicamente não funcionantes (APNF). Para isso, esse estudo analisou imagens de RM de 15 pacientes de APNF retrospectivamente separados em dois grupos: O grupo de pacientes recorrentes, definido por 7 pacientes que exibiram recorrência tumoral em um período de 4, 640 +- 0, 653 anos (média +- erro padrão) de acompanhamento clínico após a primeira abordagem cirúrgica; e o grupo de pacientes estáveis, formado por 8 pacientes com lesões consideradas estáveis em um período de 4,512 +- 0, 536 anos. Uma máscara de segmentação tridimensional da lesão tumoral foi construída manualmente por um especialista sobre a imagem 3D T1-W DCE pré-operatória para cada paciente. Em seguida, essa segmentação e a própria imagem de ressonância foram usadas para extrair 48 características numéricas de textura. Adicionalmente, 4 características clínicas foram consideradas no estudo: a imunohistoquímica, invasividade, idade na primeira cirurgia e sexo, totalizando 52 características. Cada uma destas 52 características fora testada através de testes estatísticos convencionais univariados para ver se existia evidencias do poder discriminatório dessas características para diferenciar esses dois grupos de pacientes. Mais adiante, diferentes subconjuntos dessas características foram usados para construir modelos de predição baseados na teoria de aprendizagem de máquinas (usando os algoritmos k-nearest neighbor (kNN), decision tree (DTC), e random forest (RFC)) para investigar um modelo de classificação capaz de identificar os pacientes que experimentariam recorrência tumoral após a primeira cirurgia. 9 características de textura foram consideradas individualmente significantes (p < 0, 05) na diferenciação dos grupos de paciente recorrente e estável. Afirmando esses achados, a análise com a curva ROC para cada uma das 9 características exibiu medidas de AUC de 0,803 a 0,857 significando uma boa performance de classificação. A idade, imunoistoquímica, invasividade e sexo não mostraram evidencias de associação com recorrência tumoral. As melhores performances com algoritmos de classificação foram com kNN e RFC, ambos atingiram uma especificidade de 1,000 conservando alta acurácia (0,933) e obtendo 0,991 na análise com a curva ROC, o que caracteriza uma performance de classificação quase perfeita. DTC não mostrou nenhuma melhora se comparado com os resultados das classificações univariadas. Esses resultados permitem concluir que parâmetros de textura são úteis na predição de recorrência tumoral após a primeira cirurgia em pacientes de APNF. E que os valores de predição dessas características podem ser observados por testes estatísticos univariados convencionais e por análises multivariadas através de algoritmos baseados em aprendizagem de máquinas / The present work proposes the usage of texture features computationally extracted from MRI as imaging biomarkers in the prediction of tumor recurrence in patients with non-functioning pituitary adenomas (NFPA). With this purpose, this study analyzed MR images from 15 patients of NFPA retrospectively separated in two groups: the recurrent patient group, formed by seven (7) patients who exhibited tumor recurrence in a period of 4,640 +- 0,653 years (mean +- standard error) of follow-up period after the first surgical approach; and the stable patient group formed by eight (8) patients with lesions considered stable in a period of 4,512 +- 0,536 years. A three-dimensional segmentation mask of the tumor lesion was manually performed by a specialist over preoperative 3D T1-W DCE MR image for each patient. Next, this segmentation and the preoperative MRI itself were used to extract 48 numerical textural features. Additionally, 4 clinical features were considered in the study: immunohistochemistry, invasiveness, age at first surgery, and gender, totalizing 52 features. Each one of those 52 features were tested through conventional univariate statistical tests to see if there were evidence of their discrimination power to differentiate these two patient groups. Moreover, different subsets of those features were used to build machine learning prediction models (using k-nearest neighbor (kNN), decision tree (DTC), and random forest (RFC) algorithms) to investigate a classification model capable of identifying the patients that would experience tumor recurrence after the first surgery. 9 quantitative textural features were found to be individually significant (p < 0,05) in the differentiation of recurrent and stable patient group. Affirming these findings, the ROC curve analysis for each one of those 9 features exhibited an AUC score from 0.803 to 0.857 meaning a good classification performance. Age, immunohistochemistry and invasiveness status, and gender did not show evidence of association with tumor recurrence. The best performances with classification algorithms were obtained with kNN and RFC, both reached specificity of 1.000 conserving high accuracy (0.933) and scoring 0.991 in ROC curve analysis, what characterizes an almost perfect classification performance. DTC did not show any improvement compared to the univariate classification results. These findings allow to conclude that textural features are useful in the prediction of tumor recurrence after first surgery in NFPA patients. And that the prediction value of those features can be observed with both conventional univariate statistical tests and multivariate analyses through machine learning algorithms