4 research outputs found
A machine learning framework for multi-hazards modeling and mapping in a mountainous area
No full textA Hot-Spot Analysis of Forest Roads Based on Soil Erosion and Sediment Production
No full textForest roads have been recognized as one of the significant contributors to soil erosion processes in forested areas. The construction and maintenance of forest roads can cause severe environmental impacts, including soil erosion, sedimentation, and degradation of aquatic ecosystems. The main objective of the present study is to analyze the impact of forest road networks on soil erosion and sedimentation in the context of the Zagros forestlands, Iran. This study aims to assess the soil erosion and sedimentation on forest roads in four case studies in the Zagros forestlands. This study collected data using field surveys and SEDMODL equations to determine input factors and sedimentation and erosion rates. This study found that roadside erosion is strongly correlated with geological factors, road width, and precipitation factors. The height changes of 144 benchmarks were recorded during one study year (2021–2022) on four study roads, and the measured results of erosion benchmarks indicated an average soil erosion of 3, 2.6, 4.7, and 3.5 mm per year around the Bideleh, Kohian, Nazi, and Tabarak roads, respectively. This study measured soil erosion and sedimentation at three distances (5, 15, and 25 m) from the road, and found a significant difference in the height changes of the benchmarks at varying distances from the study roads. A hot-spot analysis was conducted using GIS 10.8, and the results indicated that a significant portion of the studied forest roads had very high erosion production and hot spots. The results of the hot-spot analysis indicated that 30.8%, 22.6%, 39.8%, and 14.5% of the study forest roads, Nazi, Tabarak, Bideleh, and Kohian roads, respectively, are identified as areas with very high erosion production and hot spots. These results highlight the need for effective management strategies to minimize the impact of erosion on road infrastructure and the surrounding environment. Overall, this study provides important insights into the soil erosion and sedimentation on forest roads, and the findings presented here can be used to inform future road construction and maintenance
Neutropenia in cancer patients, risk prediction models of neutropenia, and supportive measures
No full textEpidemiology studies the causes and distribution of population health and disease conditions in defined populations. It identifies risk factors for disease which may help to prevent disease and promote health.
Each year, the American Cancer Society describes the epidemiology of cancer in the USA. Breast cancer and CLL are the most common cancers in women and adults, respectively. European data for CLL are limited. For both cancers, chemotherapy is an important treatment option. But side effects such as neutropenia and infections remain the principal dose-limiting toxicities, which may affect the effectiveness of cancer chemotherapy. Several studies evaluated risk factors for chemotherapy-induced neutropenia (CIN; absolute neutrophil count [ANC] <1.5x10^9/L) and febrile neutropenia (FN; ANC <0.5x10^9/L and oral temperature =38° for more than 1 hour): e.g. older age, recent infection, prior chemotherapy, and planned relative dose intensity greater than 85% of standard chemotherapy dosing. The prophylactic use of granulocyte colony-stimulating factors (G-CSFs) has been shown to be protective.
Based on the above mentioned risk factors, a number of risk prediction models have been developed over the years. Very often, the risk prediction models considered patient-related, tumour-related, treatment-related, or genetic factors. The majority of these models are not validated using an independent dataset. Systematic reviews of G-CSFs to prevent neutropenia are available, but do not include new long-acting G-CSFs or observational study designs.
To address the epidemiology of CLL, the incidence and risk factors of CIN and FN, and to develop and externally validate a risk prediction model for the occurrence of FN including a broad range of risk factors, three quantitative studies were conducted and published. The fourth published study summarised the efficacy, effectiveness and safety of G-CSFs for the prevention of CIN and FN.
For the first study, the author conducted a cohort analysis of the UK Clinical Practice Research Datalink (CPRD) to identify the epidemiology of CLL, the incidence of neutropenia, and changes in medical resource utilisation of CLL patients. Due to limited data regarding the incidence of neutropenia, the study focused on the epidemiology of CLL and medical resource utilisation of CLL patients. The incidence of CLL was 6.2 per 100’000 person-years and remained stable between 2006 and 2011. Medical resource utilisation in CLL patients increased over the time period from 2000 to 2012. Primary care data from the UK CPRD seemed to be valid to determine the incidence of CLL. These data may not reflect the total of medical resource use in CLL patients as chemotherapy and treatment of related complications such as infections and neutropenia are mainly performed in secondary or tertiary care.
The second study addressed the identification of risk factors and the development of a risk prediction model for FN in a hospital-based breast cancer cohort. Risk factors for FN were lower platelet count and haemoglobin, higher alanine aminotransferase (ALT), and specific allele variants of two single nucleotide polymorphisms (SNPs) in a gene involved in multidrug resistance. Genetic testing beforehand might be helpful to identify patients at a very high risk of FN. Predictive performance of the model was improved by adding genetic information but overall remained limited.
The third study used an available risk prediction model for FN in Non-Hodgkin lymphoma (NHL) patients and applied its prediction rules to an independent dataset of NHL patients. Age, weight, baseline white blood cell count, and planned chemotherapy dose were confirmed to predict the risk of FN. However, there was a decrease of the predictive performance in the independent validation dataset. This limits its use in clinical practice. But if successful risk prediction models are developed and externally validated, these may help to optimally target prophylaxis with G-CSFs to those patients at high risk of FN.
Finally, a systematic literature review was conducted to identify studies evaluating the efficacy, effectiveness and safety of G-CSFs in the prevention of CIN and FN. Most studies showed better efficacy and effectiveness for the long-acting pegfilgrastim than daily filgrastim. Efficacy and safety profiles of new long-acting G-CSFs such as lipegfilgrastim and balugrastim were comparable to pegfilgrastim. In times of increasing health care costs and scarce resources, the cost-efficient use of supportive measures is necessary.
The studies this work is based on showed that the availability of and access to appropriate data sources are necessary to develop and systematically validate risk prediction models. The findings contribute to the development of an evidence-based, efficient and cost-efficient approach to prevent neutropenia in cancer patients
Representaciones de patrones para la clasificación de datos (no-)vectoriales (no-)métricos con aplicaciones en el Monitoreo de Salud Estructural y la ingeniería de amenazas geotécnicas/naturales
No full textgráficos, tablasNowadays, data-driven modelling in structural and geo-engineering problems using Statistical Pattern Recognition and Machine Learning provides powerful and more versatile tools within a predictive framework. In contrast to the mainstream orientations of the state-of-art in data-driven structural and geo-engineering surrogates, which are based on advanced and (hyper-)parametrized classifiers, this thesis is focused on data representation issues. Firstly, for vectorial slope/landslide data, feature-based vector spaces are enriched and enhanced according to the Occam’s razor principle, which is achieved through three simple but powerful existing variants of a transparent classifier as the nearest neighbor rule. Secondly, for non-vectorial SHM data, powerful and highly discriminant dissimilarity-vector spaces are built-up using spectral/time-frequency information from structural states, adopting a
proximity-based learning scheme. In both cases, the results show the importance of a proper data representation and its key role in a bottom-up design for surrogate modelling. (Texto tomado de la fuente)Actualmente, el Reconocimiento de Patrones Estadístico y el Aprendizaje de Máquinas proveen herramientas poderosas y versátiles para el modelamiento predictivo de problemas de estructuras civiles, mecánicas y de la geo-ingeniería. A diferencia de las principales tendencias en el estado del arte en los sustitutos basados en datos en problemas de estructuras y de geo-ingeniería, esta tesis se enfoca en la representación de los datos. Primero, para datos vectoriales de taludes/deslizamientos, los espacios vectoriales basados en características son enriquecidos y mejorados de acuerdo al principio de la navaja de Occam o de parsimonia, el cual se logra mediante tres simples pero poderosos variantes ya existentes del clasificador de vecinos más cercanos. Segundo, para datos no-vectoriales pertenecientes al Monitoreo de Salud Estructural, son construidos, poderosos y altamente discriminantes, espacios de disimilitudes usando información espectral/tiempo-frecuencia, tomando un esquema de aprendizaje basado en proximidades. En ambos casos, los resultados demuestran la importancia de una apropiada representación de datos y su influencia en el diseño incremental de modelos sustitutos.DoctoradoDoctor en IngenieríaStatistical Pattern Recognition, Machine Learning and Signal ProcessingEléctrica, Electrónica, Automatización Y Telecomunicacione
