21684 research outputs found
Sort by
Integration of physical prior knowledge in machine learning imaging workflows
Get PDFIn this thesis, we introduce several scientific machine learning methods to circumvent models’ data dependencies by bridging the gap between the physical insight about the acquisition data process and the machine learning paradigms. The aim is to open neural networks’ black box by combining it with a well-known forward process, the white-box model. Having access to a well-defined white-box model, we can embed its information inside the network in order to obtain a hybrid model where we partially know how it should respond.
Focusing on medical and aerospace imaging applications, we leverage sensor calibration profile and image signal processing prior knowledge to develop three novel validation protocols via a physically faithful differentiable model. Starting from the object, through the optics, to the sensor, the entire imaging process is integrated into the machine learning workflow to detect model failures and enhance model robustness. These novel methods extend model generalization beyond classical techniques like catalogue testing or augmentation, bringing additional freedom to the data and model explainability.
Guided by the principle of metrologically precise data handling, we designed a data-centric machine learning workflow to emulate expensive satellite imaging payloads using more affordable drone image data. The emulation mimics pixel distribution and the optical properties of the target acquisition system, allowing an in silico model validation before launching the physical prototype. The experiments demonstrate the lowest resolution and signal-to-noise ratio necessary for conducting a segmentation task on satellite data, offering the optimal range of optical parameters where the model operates effectively.
While in medical imaging the acquisition process has a key role in making the model more resilient to real-world application, it does not cover the out-of-distribution negative impact on the downstream model due to missing data in sparsely annotated datasets. In this thesis, we developed a generative framework based on diffusion models for the synthesis of lung cancer tissue histological data. The model leverages the biological insight on the macroscopic cell arrangement to guide the synthesis using new features from unlabelled data. We evaluated the efficacy and fidelity of the generated content via a comprehensive data assessment and we explored the potential of synthetic data for training on in-out house data
The Wendland Movement: Anti-nuclear energy resistance in Gorleben
Get PDFAbstract not currently available
Miniaturisation and optimisation of ultrasonic scalpels for robotic surgery
No full textAbstract not currently available
Gendering political leadership in the Scottish Parliament since devolution
Get PDFThis thesis sets out to explore the extent to which the experience of political leadership in the Scottish Parliament gendered, the extent to which the Scottish Parliament is a gendered workplace and how this has gendered outcomes for political leaders. It responds to the emergent body of literature that conceptualises parliaments as gendered workplaces and how this impacts leadership. At present, this body of literature remains in its infancy and recognises that there is much to be discovered. In this light, the Scottish Parliament presents an empirically rich research setting to respond to such calls. Unlike other national parliaments, gender equality was designed in the feminist institutional blueprint of the parliament and the Scottish Parliament has enjoyed relative success on women’s representation where women have occupied 33-45% of seats since 1999. Additionally since devolution, existing literature on the Scottish Parliament has not systematically explored gender and political leadership, nor has the Scottish Parliament been conceptualised as a gendered workplace, in the 24 years of the parliament’s existence.
Empirically, this thesis is based on a unique dataset of 31 elite interviews with men and women MSPs. The cross-party interview sample includes former and current MSPs from each of the political parties elected in the first five sessions of the Scottish Parliament (1999- 2021). This thesis develops a theoretical framing by using Bourdieusian concepts of habitus, capitals, and field to understand the ways that leaders’ biographies shape their leadership practice, how leadership capitals are learned and developed, and how the rules of the field shape leadership. This is combined with a Feminist Institutionalist approach to examine the ways in which the formal and informal rules in the parliament are gendered. Using Bourdieu’s concepts, the thesis develops a distinctive leadership practice that is informed by MSPs’ pre-parliamentary histories that influence the field-specific political habitus in the parliament. Women’s inclusion has created a feminised habitus where both men and women value and embody leadership capitals, including collegiality, care, and empathy. However, conclusions drawn indicate the presence of feminine workplace ideals that exist in tension with masculine workplace practices. In sum, this thesis contributes to understandings of gender and political leadership that go beyond reductive binary gendered leadership assumptions by situating leadership in the complex political field where valued leadership capitals are informed by the ‘gendered conditions’ of the political landscape, parliamentary history, and gendered workplace rules
Novel applications of machine learning for the study of animal movement
Get PDFAnimal movement data play a crucial role in our endeavour to decode the wildlife dynamics that unfold across the Earth’s varied terrains and waterways. Collecting animal movement data involves employing a variety of modern technologies and techniques. One common method is the use of Global Positioning System (GPS) devices that are attached to animals, providing accurate location data at regular intervals. These devices allow us to track animals’ movements over time and space, giving us access to intricate details about their ranging behaviours and daily routines.
These data provide insights into migration patterns, foraging strategies, habitat preferences, and responses to environmental changes and are therefore key to an improved understanding of the behaviours, habitats and ecological interactions of various species. There are many challenges associated with the analysis of such datasets and movement ecologists have been dilligently advancing state-of-the-art statistical methods for analysing animal movement data. This effort is crucial because it enables us to extract meaningful information from the complex, large-scale datasets generated by animal tracking studies. These advanced statistical techniques help uncover hidden patterns, such as the identification of significant stopover sites during migrations or the characterisation of nuanced movement behaviours. Moreover, they facilitate the integration of environmental variables, enhancing our ability to understand how animals respond to changing landscapes and climate conditions. By refining our analytical tools, movement ecologists can provide more accurate and comprehensive insights into wildlife behaviours, aiding conservation efforts and ecological research on a global scale.
Throughout this thesis, we will talk about models of animal movement and we will give our contribution to expand the array of statistical methods that can be employed for the analysis of telemetry datasets. We will begin by reviewing some of the most commonly employed methods found in the literature. Then, we will focus on a simple one-dimensional self-propelled particle model used to simulate the dynamics of a group of locusts placed in a ring-shaped arena for an experiment, and we will leverage Gaussian processes to infer microscale properties of the group from macroscale observed variables, without deriving a formal mathematical link between the two scales, which is in many cases intractable.
Our focus will then shift to the problem of identifying different behavioural patterns from relocation data. In the fourth chapter we will describe various methods that we have conceptualised all aimed at simulating semi-Markov chains. This will be needed in the subsequent chapter, where we will introduce a flexible model scalable to large datasets that finds its applications in solving the so-called switching problems. This will be achieved by modelling the locations via an integrated OU process and by reconstructing the latent behavioural patterns via a Monte Carlo Expectation-Maximisation algorithm, where the method introduced in the previous chapter will be employed.
In Chapter 6 we will employ this method on a flock of sheep, specifically on a group level metric describing the changes in coordination of the group motion. This will enable us to reconstruct the behavioural pattern of the flock. We end the thesis with a conclusion chapter
Developing a social skills intervention for adolescents with Autism Spectrum Disorders (ASD) in Thailand
No full textBackground: The global prevalence of autism spectrum disorders (ASD) is 0.6%, similar to the prevalence of ASD in Thailand. In 2022, approximately 2.2 million Thai children were diagnosed with ASD, but only 1.16 percent obtained healthcare services through the health system. Adolescents with ASD have challenges with social communication, lack social cognition to understand or predict the behaviour of others, and lack emotional recognition to recognise the emotions of others, which negatively impacts their relationships and daily life. A number of systematic reviews have concluded that the Western-originated social skills interventions is effective in improving social skills and quality of life in adolescents with ASD. However, these interventions may not be applicable to Thailand, whose culture is different and may have different needs in, and preferences for, engaging with social skills interventions. It is essential to develop a culturally adapted social skills intervention for Thai adolescents with ASD.
Aim: The aim of this thesis was to inform the development of a culturally adapted social skills intervention for adolescents with ASD (aged 10–19 years) and their families in Thailand.
Methodology: To achieve this aim, three studies followed the initial phase of the Heuristic Cultural Adaptation framework for intervention development. A second cultural adaptation framework, the Ecological Validity Model (EVM) framework was used for evaluating the cultural domain of an evidence-based intervention and identifying the cultural needs of the stakeholders. The first study was a systematic review and meta-analysis to determine the efficacy and cultural adaptability of a social skills intervention for adolescents with ASD in Asian countries. The second study was a qualitative study which included focus groups of healthcare professionals and paired-depth interviews with adolescents with ASD and their caregivers to understand needs and perspectives towards the components and delivery formats of and necessary cultural adaptations to, a social skills programme for Thai adolescents. Lastly, given that the research was conducted during the COVID-19 pandemic, the third study was an online survey of caregivers of adolescents with ASD that investigated the impact of the COVID-19 pandemic on the experiences of Thai adolescents and their families across Thailand. This study was important to understand how the pandemic had impacted on, and provided a context for, the current landscape of social skills development in adolescents with ASD and the recommendations emerging from this PhD study for the development of a social skills programme for Thai adolescents and future research.
Results: The findings from the three studies presented in this thesis provide an understanding of the role, needs for, and cultural adaptation required in developing a social skills programme for adolescents with ASD in Thailand. The systematic review and metaanalysis (n = 15) highlighted that social skills interventions in Asian countries significantly improved social skills in adolescents with ASD. It also demonstrated that the adaptation of these interventions was crucial when adapting western intervention concepts to the local population.
The qualitative data reinforced that there is a need for social skill interventions among Thai adolescents with ASD. The data identified that the provision of a programme should assess the social skills of adolescents before giving them social skills training, teach the specific social skills, cooperate with the various learning strategies, involve parents in the programme, and encourage parents’ knowledge and skills. Furthermore, adolescents with ASD and parents were required to comprehend and practise Thai social knowledge (e.g., respect for the elderly, morality, generosity, responsibility, and Buddhism) as well as communication and behaviour techniques (e.g., assertiveness and praising) with which Thai parents were unfamiliar. Staff workload, staff competence, the impact of the COVID-19 pandemic, and stigma were identified by health professionals as potential barriers to implementing the intervention.
The online survey results confirmed that the COVID-19 pandemic, school closings, decline in outdoor activities, and postponement of treatments influenced the social skill development of Thai adolescents with ASD. However, adolescents' regular access to services, caregivers' sufficient support from the hospital, and the older age of adolescents may have helped to prevent the worsening of social skills during this time. To support the development of social skills, caregivers and adolescents with ASD required additional medical support, such as social skills training, telehealth, and home medication delivery.
Conclusion: Social skills interventions are essential for improving the ASD adolescent's social skills and quality of life. To ensure that social skills interventions appropriately engage and meet the needs of Thai adolescents with ASD and their families, current social skills interventions developed in Westernised countries need to be culturally adapted. The results from the three studies conducted here provide recommendations for informing the aims, cultural adaptation, and delivery of a social skills intervention for Thai adolescents and their families. The outcome of a culturally adapted intervention may help to increase engagement with and participation in the programme and improve outcomes for adolescents with ASD in Thailand. It is recommended that future research involves the testing of an adapted social skills intervention with regards to these outcomes
Optimisation of radiotherapy for pancreatic cancer
Get PDFBackground:
Pancreatic ductal adenocarcinoma (PDAC) remains a cancer of unmet need, with modest improvement in overall survival (OS) being realised over the past few decades. State of the art RT technology has advanced treatment protocols for PDAC, increasing the ability to deliver sophisticated RT that exploits the therapeutic ratio and overcomes the innate radioresistance. PDAC is a disease that requires optimised multi-modality protocols. This can facilitate delivery of dose escalated radiotherapy (RT) of high-quality that can be investigated in studies to determine if clear benefits can be established.
Uncertainties continue to be problematic in the planning and delivery of RT for PDAC, with many challenges present in the multi-factorial chain. By considering the relationship between pathway components and the impact they have on the delivery of safe, precise, and accurate treatment, RT can be improved.
Background Pancreatic ductal adenocarcinoma (PDAC) remains a cancer of unmet need, with modest improvement in overall survival (OS) being realised over the past few decades. State of the art RT technology has advanced treatment protocols for PDAC, increasing the ability to deliver sophisticated RT that exploits the therapeutic ratio and overcomes the innate radioresistance. PDAC is a disease that requires optimised multi-modality protocols. This can facilitate delivery of dose escalated radiotherapy (RT) of high-quality that can be investigated in studies to determine if clear benefits can be established.
Aim:
To optimise components of the RT planning and delivery pathway using a linear accelerator (linac)-based treatment platform to improve safety, accuracy, and precision of stereotactic ablative RT for PDAC.
Methods:
I. A retrospective study evaluated dosimetric and clinical outcomes for patients treated for PDAC with conventionally fractionated VMAT. This was to describe clinical outcomes with standard of care treatment, which would provide a baseline measurement.
II. A survey methodology was used to understand the views of clinical oncologists (CO) and clinical oncology trainees (COtrain), to determine the importance and priorities of areas that required optimisation of PDAC RT.
III. The volumetric impact of introducing MR-CT fusion to RT planning for pancreatic cancer was evaluated for CO and COtrain delineations. Volumes were quantified with and with and without MR, and concordance of volumes were compared to gold standard (GS) volumes.
IV. Image quality scoring criteria was developed and used to evaluate if breath-hold (BH) imaging could improve subjective image quality, visualisation of structures and confidence in decision making, using a comparison of CBCT images acquired in exhale breath-hold (CBCT_EBH) and free-breathing (CBCT_FB). This was carried out for expert (EXP) and clinical observers (OBS).
V. Planning and delivery of stereotactic ablative RT (SABR) to PDAC patients was compared using two methodologies. These were i) a motion encompassing technique which used 4DCT in free-breathing (4DCT_FB) to plan with an ITV approach, with CBCT_FB verification; and an individualised EBH technique using a 3D-contrast enhanced CT in EBH (3DCECT_EBH) data set and the exhale phase of the 4DCT acquired in freebreathing (FB) (4DCT_EXHALE) for planning, with CBCT_EBH verification.
Results:
Thirty-six patients were evaluated with clinically acceptable plans being created and delivered successfully for all patients. The maximum acute toxicity experienced was a grade 2 for anorexia (3 patients), diarrhoea (1 patient), nausea (6 patients) and pain (2 patients). The median OS for potentially operable patients was 16 months (95%CI 11-27) and for inoperable was 14 months (95%CI 6-17).
CO and COtrain respondents described PDAC delineation was challenging, and they agreed that optimisation of the pathway was important. Optimisation of planning images; peer review of volumes and plans with a multi-disciplinary team (MDT); and improving tumour delineation by reducing inter-observer variability (IOV) were ranked as the most important.
Volumes of GS structures were smaller than mean observer structures. Across the population, MR volumes were smaller than those created with CT only (GTV_3D). The highest magnitude of mean difference was observed directly between volumes with and without MR. Using MR and CT for delineation showed an increase in dice-similarity coefficient (DSC).
Image quality scores were improved for EXP and OBS in CBCT_EBH images and was statistically significant. An improvement in mean standard deviation (SD) scores for celiac artery (CA), superior mesenteric artery (SMA) and superior mesenteric vessel (SMV), duodenum, GTV/ITV and PTV were shown for EXP and OBS. Increased confidence was observed for the EXP and OBS groups.
Clinically acceptable plans were created using both methodologies. Similar planning target volume (PTV) coverage were achieved for PTV_FB and PTV_BH, with OAR dose reduction being shown in the PTV_BH plans. CBCT_EBH significantly improved image quality, visualisation of structures and confidence in decision making. The assessment of PTV_BH resulted in 97% of scores, there was confidence in verifying coverage, significantly more than in CBCT_FB at 26%.
Conclusions:
In conclusion, the evaluation of thirty-six patients in the initial retrospective study showed that VMAT plans resulted in acceptable acute toxicity. No grade >3 toxicity reported, and grade 2 symptoms being reported in a small percentage of cases. The median overall survival for both potentially operable and inoperable patients were reported. This formed a baseline for conducting the next stages of the thesis.
The survey carried out highlighted the challenges in delineating pancreatic cancer (PDAC) and emphasised the collective desire to optimise the pathway when developing SABR. Various strategies, including peer review, improved imaging, and reducing inter-observer variability, were identified as crucial in this context.
The delineation study demonstrated that IOV was high across volumes using MR alone, CT alone, and registered images. MR imaging alone resulted in smaller target volumes, but more agreement was shown in volumes delineated with an ITV approach which used MR and CT.
CBCT image quality scores assessed using scoring criteria developed within this thesis showed improvements with CBCT_EBH. This was shown for EXP and OBS groups when compared to CBCT_FB scores. This enhanced visualisation of structures and improved confidence in decision-making.
A comparison of plans for PTV_FB and PTV_EBH showed both methodologies were effective in generating clinically acceptable plans, with BH resulting in similar PTV outcomes and reduced dose to normal tissue. A clear and important advantage of a BH SABR technique was that CBCT_EBH offered superior image quality and increased confidence in verifying target coverage. This demonstrated that CBCT_EBH is superior to CBCT_FB in delivering PDAC SABR and should be considered a necessary method for linac-based RT.
This thesis has investigated a number of processes within the PDAC SABR pathway, identifying uncertainties which can be minimised, allowing protocols to be optimised. These findings contribute to the ongoing effort to optimise techniques that can be implemented in clinical trials going forward, with an aim to improve outcomes for pancreatic cancer patients
Gas phase pyrolytic approach for stable isotope analysis of hopanoids lipids
Get PDFMethanotrophs are known for their role in mitigating CH4 emissions, with aerobic methanotrophs adapted to a wide range of environments. In oxic settings, they are able to utilise this high global warming potential molecule as a source of energy, oxidising it to CO2, or as a source of carbon, incorporating the CH4-carbon’s (C-CH4) isotopic signature into the intramolecular structure of the membrane-located hopanoid lipids. Being decay-resistant, hopanoids can be used to study past CH4 cycles.
Compound-specific isotope analysis (CSIA) of diploptene, a hopanoid biomarker, has been initially employed in this work to identify the advantages and limitations of this technique to assess past CH4 consumption within three peatlands of Pastaza-Marañón Foreland Basin (PMFB). This study thus provides the first multi-biomarker, yet low-resolution, isotopic reconstruction of the three peatlands, with CSIA values of hopanoids, as well as of n-alkanes and fatty acids interpreted in conjunction with previously measured biomarker concentrations and a wide range of palaeoenvironmental data from the same peatlands or area. Isotopic values of n-alkanes and diploptene reflect the CH4-producing conditions of the area, fluctuating with past changes in each peatland’s history. It was however found that methanotrophs were not the sole producers of hopanoids and, due to source dilution and the fact that the three aerobic methanotroph types can also incorporate different amounts of CCO2, via distinctive pathways, diploptene’s methanotrophic isotopic signature can become muted in these (palaeo)environmental samples at the CSIA level.
These limitations, together with low compound concentrations and co-elution, were also encountered in simplified, one-diploptene source methanotrophic lab cultures conducted in this work to study diploptene’s CSIA evolution in relationship to isotopically known CH4 and CO2 sources. Even when the bio-producer is known, diploptene’s CSIA was influenced by the varying quantities of C-CH4 and C-CO2 incorporated by type I and II methanotrophs. New tools are thus required to assess methanotrophic processes in natural environments and increase confidence in source attribution. This work develops theoretical and analytical methodologies for the gaseous pyrolysis of diploptene that could enable the detection of methanotrophy at the intramolecular (or position-specific) level through isotopic analyses.
Analytical instruments capable of accessing this information are currently restricting these new isotopic applications through complex setups, molecule-specific uses and high costs. Gaseous pyrolysis is demonstrated in this thesis on a bespoke GC-FID instrument (prep-GC) adapted to overcome these limitations. The development of the prep-GC instrument and methodology took place in a step-like fashion, with experiments aimed at increasing and testing its capacities on compounds such as methanol, MTBE, 2,3 DMN, squalene and diploptene, and their multiple pyrolysates. The prep-GC is demonstrated to be able to analyse compounds with molecular masses between 32 - 410 g/mol and at least up to 850℃ pyrolytic temperatures, making it a universal analytical instrument for the pyrolysis of volatile and semi-volatile compounds. Experiments presented in this thesis demonstrate the ability of the prep-GC to heart-cut and concentrate, within a trap, compounds of interest, release, pyrolyse and transfer the pyrolysates to a second detector for quantification and identification. The analytical pyrolytic methodologies were further adapted and the results were confirmed through in-silico pyrolysis, using Reaction Mechanism Generator (RMG, Gao et al., 2016).
Diploptene, and hopanoids in general, have never been subjected to pyrolysis to investigate if the C-CH4 moieties can be accessed and their relationship to the CH4 cycle. The pyrolytic breakdown mechanism of diploptene was assessed first on RMG and the results were experimentally studied. On RMG, for the simulated pyrolytic temperature of 750℃, isoprene was the main pyrolytic fragment, produced from squalene, which occurs as an unstable intermediate. During pyrolysis, the diploptene backbone therefore unfolded, undoing the action of the squalene-hopene cyclase, and was broken down into isoprene units. Theoretical and experimental evidence of diploptene, squalene and isoprene pyrolysis are presented to support isoprene as the main diploptene pyrolysis fragment between 550℃-625℃. Isoprene however is demonstrated to not provide new isotopic evidence when compared to diploptene, as it does not allow a direct isotopic investigation of the C-CH4 and C-CO2 incorporated through the different methanotroph bio-cycles. Isoprene ozonolysis is proposed as a future tool that could access this intramolecular information, post diploptene (and squalene) pyrolysis. Finally, although produced from the same predecessor (squalene), RMG simulations indicate that other hopanoids present different breakdown mechanisms, requiring individual breakdown pyrolytic assessments for future PSIA studies.
The development of the prep-GC instrument for gaseous pyrolysis biomarkers at low environmental concentrations and the exploration of hopanoid pyrolysis mechanisms in this thesis contribute to advancing the understanding of methanotrophy and its detection via isotopic studies, paving the way for new intramolecular isotopic investigations
Performance and commissioning of the BigBite Timing Hodoscope for nucleon form factor measurements at Jefferson Lab
Get PDFThe BigBite Timing Hodoscope detector is the primary subject of this thesis. The Super BigBite Spectrometer is a Jefferson Lab Hall A Collaboration project that has and will continue to measure nucleon electromagnetic form factors. This spectrometer includes the Timing Hodoscope which provides high resolution particle timing data for scattered electrons in the electron arm of BigBite. The Timing Hodoscope utilizes 90, 25 × 25 × 600 mm3 scintillator bars stacked on top of each other to form a single detector plane, and these bars are connected to 180 photo-multiplier tubes via light guides. Particles collide with the scintillating material creating a shower of optical photons and these particle events in the bars are collected to generate signals that are readout by the data acquisition (DAQ) electronics. NINO ASIC amplifier-discriminator cards output signals from the photo-multiplier tubes into analogue and logic signals, which are sent to analogue-to-digital (ADC) and time-to-digital (TDC) converter data acquisition readout modules. This data is then used for analysis of the detector.
The focus of this thesis is the construction, commissioning, calibration, and performance of the BigBite Timing Hodoscope before and during the first of five nucleon electromagnetic form factor experiments at Jefferson Lab Hall A. Before the neutron magnetic form factor, G n M, experiment, cosmic ray data was collected during commissioning to confirm proper operation of the Timing Hodoscope electronics by observing the ADC and TDC data. Commissioning studies for charge normalization, gain matching, and other ADC and TDC detector data variables were performed before moving the detector into Hall A. Following installation in Hall A, several calibration studies were implemented to fine-tune the detector in preparation for use in the experiment. The calibration studies included analysis of timing cuts, TDC alignment, the time-walk effect, time difference offsets, and scintillator velocity corrections. Once the Timing Hodoscope was well-calibrated, data-taking during the experiment commenced and the beamon-target data was used to characterize the Timing Hodoscope performance during the G n M experiment run-time. The performance analysis included studies observing energy deposit, cluster size, rates, accidentals, pile-up, tracking efficiency, position resolution, and time resolution. After application of physics cuts to ensure a data set comprised of particle tracks corresponding to elastic electrons, which is the main data of interest for measurement of G n M, the Timing Hodoscope is shown on average across all kinematic settings to have a >98% tracking efficiency, a position resolution of 4-6 cm in the non-dispersive plane and 1.5-2 cm in the dispersive plane, and a time resolution of 500-750 ps. These performance results are compared to a GEANT4 based performance simulation of the BigBite Timing Hodoscope for reference, showing to what degree the measured performance values match those taken from the simulation
Data sources:
Access Repository Dashboard
Do you manage Open Research Online? Become a CORE Member to access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard! 👇