Дисертації з теми "Deep vessels"

This master's thesis deals with the classification of the retinal blood vessels in retinal image data. The thesis contains a description of anatomy of the human eye with focus on the blood circulation, and imaging and diagnostic methods of the retina are briefly mentioned further. The thesis also summarizes methods of the blood circulation classification with emphasis on the deep learning. The practical section was implemented in Python programming language and describes the pre-processing of the data with determination of AV ratio. Based on a literature search, the U-net architecture was chosen for the classification of the retinal blood vessels. The architecture was modified using the open-source Keras library and tested on images from the experimental video-ophthalmoscope. The modified architecture was initially used for classification of vessels into the corresponding classes and because of unsatisfying results was modified another architecture segmenting retinal vessels, arteries or veins and a proposition of a method of the blood vessels classification.

The aim of the following thesis was to study the issue of optical disc and retinal vessels segmentation in ophthalmologic sequences. The theoretical part of the thesis summarizes the principles of different approaches in the field of deep learning, which are used in connection with the given issue. Based on the theoretical part, methods for optical disk segmentation and retinal vessel segmentation based on the convolutional neural networks Linknet, PSPNet, Unet and MaskRCNN are proposed. The practical part of the thesis deals with the description of their implementation and subsequent evaluation.

Introduction: The DIEP flap is a popular choice for breast reconstruction, though selection of which perforating blood vessel(s) to supply the flap is still largely based on surgeon preference, with little evidence to support numbers or location of perforators. In addition, many surgeons routinely discard zone IV of the flap, limiting the size of transferrable tissue. The aim of this research was to investigate the effect of number and location of perforators within a DIEP flap, on the total pedicle flow and perfusion of zone IV fat and skin. Methods: This research comprised of two studies; an animal model and a patient study: 1) 20 cranially-based abdominal epigastric perforator flaps were raised in Wistar rats on two perforators. The perforators were sequentially clamped and released in a randomised order and total pedicle flow (measured using microvascular flow-probes) and skin perfusion (measured using laser Doppler Flowmetry) was recorded on the following perforator combinations: • P1 (superior perforator) • P2 (inferior perforator) • P1+2 (both perforators) In addition, half of the animal flaps were randomised to receive a single (15 minute) period of pedicle-clamped ischaemic preconditioning after raising, with all measurements repeated to observe any effect. 2) 13 DIEP flaps were raised in post-mastectomy patients requiring breast reconstruction on two perforators. These were clamped and released as before to assess perfusion of fat and skin in zone IV using SPY Indocyanine-green-fluorescence-angiography scans on the same perforator combinations as in our animal study, listed above. Results: All data were analysed using non-parametric analyses and revealed that in our animal model, total pedicle flow was significantly (p<0.001) greater on a single perforator compared to two but no significant differences were identified in the flap skin perfusion. In our clinical study a single superior perforator supplied zone IV significantly (p=0.039) better than both peroforators, though this was not observed with the single inferior perforator. No significant differences were seen in zone IV skin perfusion. A single period of ischaemic preconditioning significantly (p<0.05) increased the total pedicle flow, but not the skin perfusion in our rat model. Conclusions: Possible reasons for these observed differences could be related to the flow dynamics and resistances specific to perforator flap anatomy and physiology and the possibility of vessel shunting in the subcutis.

A major limitation in tissue engineering is the lack of nondestructive methods to assess the development of tissue scaffolds undergoing preconditioning in bioreactors. Due to significant optical scattering in most scaffolding materials, current microscope-based imaging methods cannot â seeâ through thick and optically opaque tissue constructs. To address this deficiency, we developed a scanning fiber imaging method capable of nondestructive imaging of fluorescently labeled cells through a thick and optically opaque vascular scaffold, contained in a bioreactor. This imaging modality is based on local excitation of fluorescent cells, acquisition of fluorescence through the scaffold, and fluorescence mapping based on the position of the excitation light. To evaluate the capability and accuracy of the imaging system, human endothelial cells, stably expressing green fluorescent protein (GFP), were imaged through a fibrous scaffold. Without sacrificing the scaffolds, we nondestructively visualized the distribution of GFP-labeled endothelial cells on the luminal surface through a ~500 µm thick tubular scaffold at cell-level resolutions and distinct localization. These results were similar to control images obtained using an optical microscope with direct line-of-sight access. Through a detailed quantitative analysis, we demonstrated that this method achieved a resolution of the order of 20-30 µm, with 10% or less deviation from standard optical microscopy. Furthermore, we demonstrated that the penetration depth of the imaging method exceeded that of confocal laser scanning microscopy by more than a factor of 2. Our imaging method also possesses a working distance (up to 8 cm) much longer than that of a standard confocal microscopy system, which can significantly facilitate bioreactor integration. This method will enable nondestructive monitoring of endothelial cells seeded on the lumen of a tissue-engineered vascular graft during preconditioning in vitro, as well as for other tissue-engineered constructs in the future.
Ph. D.

A system for vessel characterization aimed at detecting deep vein thrombosis (DVT) in the lower limbs has been developed and evaluated using ultrasound image processing, location and force sensors measurements, blood flow information and a protocol based on the current clinical standard, compression ultrasound. The goal is to provide an objective and repeatable system to measure DVT in a rapid and standardized manner, as this has been suggested in the literature as an approach to improve overall detection of the disease. The system uses a spatial Kalman filter-based algorithm with an elliptical model in the measurement equation to detect vessel contours in transverse ultrasound images and estimate ellipse parameters, and temporal constant velocity Kalman filters for tracking vessel location in real-time. The vessel characterization also comprises building a 3-D vessel model and performing compression and blood flow assessments to calculate measures that indicate the possibility of DVT in a vessel. A user interface designed for assessing a vessel for DVT was also developed. The system and components were implemented and tested in simulations, laboratory settings, and clinical settings. Contour detection results are good, with mean and rms errors ranging from 1.47-3.64 and 3.69-9.67 pixels, respectively, in simulated and patient images, and parameter estimation errors of 5%. Experiments showed errors of 3-5 pixels for the tracking approaches. The measures for DVT were evaluated, independently and integrated in the system. The complete system was evaluated, with sensitivity of 67-100% and specificity of 50-89.5%. System learnability and memorability were evaluated in a separate user study, with good results. Contributions include a segmentation approach using a full parameter ellipse model in an extended Kalman filter, incorporating multiple measurements, an alternate sampling method for faster parameter convergence and application-specific initialization, and a tracking approach that includes a sub-sampled sum of absolutes similarity calculation and a method to detect vessel bifurcations using flow data. Further contributions include an integrated system for DVT detection that can combine ultrasound B-mode, colour flow and elastography images for vessel characterization, a system interface design focusing on usability that was evaluated with medical professionals, and system evaluations through multiple patient studies.

La malattia di Alzheimer (AD) è un disordine neurodegenerativo che rappresenta la forma più comune di demenza negli adulti sopra i 65 anni, mentre la compromissione cognitiva lieve (MCI) è una condizione che in alcuni casi può rappresentare una fase prodromica della malattia di Alzheimer, mentre in altri, è comune in pazienti con la malattia dei piccoli vasi cerebrali (SVD). In questo elaborato sono state sviluppate due reti neurali convoluzionali 2-D, NeuroNet-1 e NeuroNet-2 (o NeuroNet), ed applicate alla classificazione a 2 vie di: a) MCI con SVD (40 pazienti in totale) con dati di diffusione come gli indici di diffusività media (MD), anisotropia frazionaria (FA) e moda del tensore di diffusione (MO); b) AD (200 pazienti in totale) con dati MRI T1-pesati. NeuroNet-2 è basata su NeuroNet-1, attraverso considerazioni frutto di una fase preliminare di studio. I risultati sui dati di diffusione suggeriscono che, l'utilizzo di un approccio multi-modalità e di un numero di fette analizzate in relazione al numero di soggetti, comportino risultati migliori. Infatti, l'accuratezza sui dati di test ottenuta nello studio multi-modalità su 6 fette è di 0.97±0.08 (media±deviazione standard). Inoltre, l'utilizzo di tecniche di interpretazione dell'apprendimento, come Gradient-weighted Class Activation Mapping (Grad-CAM) e test di occlusione, ha permesso di valutare le regioni cerebrali più importanti nella predizione e dimostrare che, oltre ai lobi frontali, alcune regioni di sostanza bianca come il corpo calloso, il tapetum, le radiazioni talamiche posteriori ed il braccio anteriore della capsula interna, sono molto importanti nella predizione delle funzioni esecutive. Infine, nello studio che riporta risultati migliori, l'ordinamento secondo importanza delle mappe di indici di diffusione considerate risulta MD>MO>FA. Invece, sui dati MRI T1-pesati è stata ottenuta un'accuratezza sui dati di test di 0.97±0.01.

The main goal of this dissertation is to study and analyze different approaches based on deep learning techniques for the segmentation of retinal blood vessels. In order to do so, different design and architectures of CNN's will be studied and analysed, as their results and performance are evaluated and compared with the available algorithms. One other important objective of this work is to study and evaluate the different techniques that have been used for vessel segmentation, such as machine learning, and how these can be combined with the deep learning approaches. By Analyzing the features that the learned models are using to perform classification and combining them with different machine learning techniques (such as Random Forest and SVM Classifiers), another goal is to proposed a solution or set of solutions to perform the retinal vessel segmentation.

Dissertação de mestrado integrado em Engenharia Biomédica (área de especialização em Eletrónica Médica)
Medical image segmentation plays a crucial role in the medical field, since it allows performing quantitative analyses used for screening, monitoring and planning the treatment of numerous pathologies. Manual segmentation is time-consuming and prone to inter-rater variability. Thus, several automatic approaches have been proposed for medical image segmentation and most are based on Deep Learning. These approaches became specially relevant after the development of the Fully Convolutional Network. In this method, the fully-connected layers were eliminated and upsampling layers were incorporated, allowing one image to be segmented at once. Nowadays, the developed architectures are based on the FCN, being U-Net one of the most popular. The aim of this dissertation is to study Deep Learning architectures for medical image segmentation. Two challenging and very distinct tasks were selected, namely, retinal vessel segmentation from retinal fundus images and brain tumor segmentation from MRI images. The architectures studied in this work are based on the U-Net, due to high performances obtained in multiple medical segmentation tasks. The models developed for retinal vessel and brain tumor segmentation were tested in publicly available databases, DRIVE and BRATS 2017, respectively. Several studies were performed for the first segmentation task, namely, comparison of downsampling operations, replacement of a downsampling step with dilated convolutions, incorporation of a RNN-based layer and application of test time data augmentation techniques. In the second segmentation task, three modifications were evaluated, specifically, the incorporation of long skip connections, the substitution of standard convolutions with dilated convolutions and the replacement of a downsampling step with dilated convolutions. Regarding retinal vessel segmentation, the final approach achieved accuracy, sensitivity and AUC of 0.9575, 0.7938 and 0.9804, respectively. This approach consists on a U-Net, containing one strided convolution as downsampling step and dilated convolutions with dilation rate of 3, followed by a test time data augmentation technique, performed by a ConvLSTM. Regarding brain tumor segmentation, the proposed approach achieved Dice of 0.8944, 0.8051 and 0.7353 and HD95 of 6.79, 8.34 and 4.76 for complete, core and enhanced regions, respectively. The final method consists on a DLA architecture with a long skip connection and dilated convolutions with dilation rate of 2. For both tasks, the proposed approach is competitive with state-of-the-art methods.
A segmentação de imagens médicas desempenha um papel fundamental na área médica, pois permite realizar análises quantitativas usadas no rastreio, monitorização e planeamento do tratamento de inúmeras patologias. A segmentação manual é demorada e varia consoante o técnico. Assim, diversas abordagens automáticas têm sido propostas para a segmentação de imagens médicas e a maioria é baseada em Deep Learning. Estas abordagens tornaram-se especialmente relevantes após o desenvolvimento da Fully Convolutional Network. Neste método, as camadas totalmente ligadas foram eliminadas e foram incorporadas camadas de upsampling, permitindo que uma imagem seja segmentada de uma só vez. Atualmente, as arquiteturas desenvolvidas baseiam-se na FCN, sendo a U-Net uma das mais populares. O objetivo desta dissertação é estudar arquiteturas de Deep Learning para a segmentação de imagens médicas. Foram selecionadas duas tarefas desafiantes e muito distintas, a segmentação de vasos retinianos a partir de imagens do fundo da retina e a segmentação de tumores cerebrais a partir de imagens de MRI. As arquiteturas estudadas neste trabalho são baseadas na U-Net, devido às elevadas performances que esta obteve em diversas tarefas de segmentação médica. Os modelos desenvolvidos para segmentação de vasos retinianos e de tumores cerebrais foram testados em bases de dados públicas, DRIVE and BRATS 2017, respetivamente. Vários estudos foram realizados para a primeira tarefa, nomeadamente, comparação de operações de downsampling, substituição de uma camada de downsampling por convoluções dilatadas, incorporação de uma camada composta por RNNs e aplicação de técnicas de aumento de dados na fase de teste. Na segunda tarefa, três modificações foram avaliadas, a incorporação de long skip connections, a substituição de convoluções standard por convoluções dilatadas e a substituição de uma camada de downsampling por convoluções dilatadas. Quanto à segmentação de vasos retinianos, a abordagem final obteve accuracy, sensibilidade e AUC de 0.9575, 0.7938 e 0.9804, respetivamente. Esta abordagem consiste numa U-Net, que contém uma convolução strided como operação de downsampling e convoluções dilatadas com dilation rate de 3, seguida de uma técnica de aumento de dados em fase de teste, executada por uma ConvLSTM. Em relação à segmentação de tumores cerebrais, a bordagem proposta obteve Dice de 0.8944, 0.8051 e 0.7353 e HD95 de 6.79, 8.34 e 4.76 para o tumor completo, região central e região contrastante, respetivamente. O método final consiste numa arquitetura DLA com uma long skip connection e convoluções dilatadas com dilation rate de 2. As duas abordagens são competitivas com os métodos do estado da arte.

碩士
國立交通大學
生醫工程研究所
108
Cardiovascular disease is increasing as a major healthcare burden year by year in the aging industrial society. It is also consistently among the top ten causes of death for many years. Its early prevention or treatment is a major issue at personal and social levels. The uncertainty of risk in invasive approaches such as surgery and stents are usually not preferred for patients. Non-invasive treatments that can alleviate symptoms or even cure are more acceptable to the patients, especially in the early stage. In the early 1960s, Harvard University pioneered the treatment mechanism of Enhanced External Counterpulsation (EECP). The current EECP equipment inflates and deflates the air in cuffs in the inflatable pressure pants and applied pressure to the legs and synchronized with the heart relaxation, pushing more blood back to the heart. EECP has been confirmed that it can significantly increase blood reperfusion, improve blood circulation, and improve the situation of ischemia by The United States Food and Drug Administration (FDA). This study aims to develop a heart-relaxation synchronized deep vessel compression system with a monitoring system. It can detect the state of the heart through Photoplethysmogram (PPG) and synchronously compress the deep blood vessels of the lower extremities according to the diastolic sequence of the heartbeats to increase blood reperfusion by improving blood circulation, thus assisting patients with heart failure and improving patients' self-mobility. A mechanical structure will be able to squeeze to and fro the lower extremities by the roller on the structure which is locked up by a belt-type linear actuator. With this device, we can complete the external counter-pulsation, achieving the same effect as the inflatable cuffs for segment compression. Not only reducing the cost of the instrument but also shrinking the size of the system. It can be popularized in the hospital to have more than one device with its increased mobility, therefore can be used directly on the bedsides, and can benefit more patients and save costs. Through this study, we expect to be able to develop a device to fit into small clinics and even for personal household applications in order to prevent and treat heart failure and other potential cardiovascular diseases.

碩士
國立成功大學
系統及船舶機電工程學系碩博士班
92
In this paper, we will find and proof the formulas about the crushing pressure of cylindrical pressure vessel by using the result of experiment. The experimental equipment is the experimental cabin of deep-sea pressure simulation made in Taiwan. It can work in 120MPa which is respectably 1200 meters-deep. 　　The experiment is dividing into two parts which are crushing experiment of flawing and non-Flawing cylindrical pressure vessel. In the part of non-Flawing cylindrical pressure vessel, we will perform crushing experiment by using various sizes of cylinders. And we will find the formulas fitting for this experiment by getting the experimental values of crushing pressure. In the part of flawing cylindrical pressure vessel, we will use the crushing pressure values of various flaws to find the major factor effects the crush characteristics. 　　Finally, we can estimate the crushing pressure of flawing cylindrical pressure vessel by combining the crushing formulas of non-Flawing cylindrical pressure vessel with the experimental data of flawing cylindrical pressure vessel.

Ultrasound imaging techniques have become increasingly successful in the medical field as they provide relatively low cost and totally safe diagnosis. Doppler methods focus on blood flow for the diagnosis and follow-up of cardiovascular diseases. First Doppler methods only measured the axial component of the motion. More recently, advanced methods have solved this problem, by estimating two or even all three velocity components. In this context, high frame rate (HFR) imaging techniques, based on the transmission of plane waves (PW), lead to the reconstruction of 2-D and 3-D vector maps of blood velocity distribution. The aim of this Ph.D. project was to develop novel acquisition schemes and processing methods for advanced ultrasound Doppler systems. Each development step was based on simulations and experimental tests. Simulations were based on Field II©, while experiments were conducted by using the ULA OP 256 open scanner. In particular, the recently proposed 2-D HFR vector flow imaging (VFI) method (DOI: 10.1109/TUFFC.2014.3064), based on the frequency domain for displacement estimation, was thoroughly investigated. Three main issues were addressed: the high underestimation of blood flow velocity observed when examining vessels at great depths, the high computational load, which hindered any real-time implementation and the lack of information about the third velocity component. Specifically, the progressive broadening of the transmitted beam on the elevation plane due to the acoustic lens was demonstrated to be responsible for the underestimation. The computational cost was reduced by processing demodulated and down-sampled baseband data instead of radiofrequency data, and a preliminary real time version of the 2-D VFI method was implemented. It was also found that a more efficient implementation could be obtained by exploiting parallel computing and graphic processing units (GPUs). An expansion circuit board for the ULA-OP 256 hardware, which allocates GPU resources, was thus designed and built. This new system architecture may allow the implementation of even more complex algorithms, such as the 3-D VFI methods. In particular, it will be possible to implement the novel method for 3D VFI that was developed and tested during this Ph.D. project. Such method suitably extended the 2D VFI approach by proposing an efficient estimation strategy that considerably limits the overall computational load.

碩士
國立臺灣大學
農業經濟學研究所
106
The value of far sea fisheries is increasing significantly and the purse-seine ships are the important wealth-making equipment. Because ships under long-term sailing in the sea have to face the danger or unexpected accidents, it is easy for ships to cause serious depletion problems. Therefore, ship repair and maintenance provide ship’s seaworthiness and security. As a result, ship docking repair is important and the analysis of docking repair costs can provide cost-control effectively. The seven 1,100-ton purse-seine ships from the M Company are analyzed by the study during the period from 2006 to 2015. The costs and expenses incurred during the docking period were analyzed and compared for each three-year ship to be docked based on the age,cost and expense. The findings found that the crew on weekdays for the maintenance of ships do not fulfill their duties, making the ship failure rate,docking technology is not accurate, or parts are not enough to use, are caused by the increase in costs. Therefore, the study suggests that in the case of ship docking, the M Company strengthen the training os seafarers’ skills, improve the maintenance and maintenance of the crew, and upgrade the docking technology to achieve effective cost control and improve productivity.

碩士
國立屏東科技大學
機械工程系所
95
Demand of strength, anti-corrosion, and pressure resistance for pressure vessel is gradually increased, thus the pressure vessel end closure (PVEC) of clad head plate manufactured by the spinning process of double-layer clad sheet is successfully explored and developed. Using the spinning process to form the clad head plate is suitable for the large dimension; however the small dimension and thin clad head plate can be adopted by the deep drawing process. In this study, the clad head plate is formed by the deep drawing process through the die design and analysis based on the SUPERFORM software. It is suitable for the small dimension and thin clad head plate forming, and the precision and dimension can be controlled. In this study, the analysis of deep drawing using the FEM SUPERFORM is carried out. The various equivalent stress-strain distributions, forming forces, shape and dimension of clad head plate after the drawing are explored to realize the effects of different parameters on the whole forming process. For realizing the SUPERFORM application to the applicability of PVEC in deep drawing, adopting the experiment of deep drawing in other references so as to verify the exactness of forming simulation of single-layer head plate. Throughout this study, the effects of the blank holder gap, the radii of die and punch, the gap between punch and die, the frictional coefficient ratio etc upon the maximum forming force and the reduction ratio of thickness, are discussed systematically.

碩士
國立臺灣海洋大學
電機工程學系
107
The main purpose of this thesis is to construct a system for detecting the location of small blood vessel lesions in the brain based on brain MRI images and a deep learning method. We hope to provide faster and more accurate lesion information to assist physicians in making decisions as making brain diagnoses. Nowadays, doctors need to use the images of Computed Tomogram (CT) or Magnetic Resonance Imaging (MRI) to make the most judgment based on the brain diagnosis. Each image must be processed by the physician. Finding the size and location of the lesion can be used for diagnosis. Therefore, this paper uses the architecture of the convolutional neural network to analyze the brain vascular lesions by analyzing brain medical images, and to mark the found blocks medically, and finally output a new image through the system. The image will show the contents of the label, including the location and contour of the lesion. Our proposed method can reduce the labor and time required for the doctor to mark the label. Auxiliary tools are adopted to review and reduce the chance of misjudgment, and it provides the physician with the information needed to diagnose the brain for the patient. The detection of small vessel lesions in the brain of this system has been able to successfully identify more than 80% of the lesions, and the detected lesions are also about 80% in terms of positioning. The contour of the lesions is also labeled with the physician. The appearance is also more than 70% similar. Most of the parts that are not found or misjudged belong to the very small part of the label, and there is no over-labeling problem. These experimental data also prove the method used in this paper. It is effective and feasible for the medical problem of detecting small vessel lesions in the brain.

碩士
國立臺灣海洋大學
海洋資源管理研究所
95
At the 19th Regular Meeting of the International Commission for the Conservation of Atlantic Tunas (hereinafter referred to as ICCAT) held in November 2005, the Commission adopted Recommendation 05-02, which requires the number of fishing vessels under Chinese Taipei’s registry authorized to conduct a directed fishery for bigeye tuna and albacore tuna in the Convention area shall be limited to no more than 15 and 60 respectively in 2006. Besides, the catch limit of Atlantic bigeye tuna for Chinese Taipei in 2006 shall also be limited at the level of 4,600 tons. In addition, the Commission also asks Chinese Taipei for taking actions in compliance with the additional requirements in the Attachment of the Resolution, including vessel reduction by scrapping 160 fishing vessels grater than 24 meters and strengthening the management of small scale fishing vessels. This research is aiming to explore the reasons why ICCAT determinately decided to adopt the Resolution against Taiwan, and to explore the extent of effect on Taiwan tuna fishing industry that the Resolution has caused. For being consistent with the Resolution, only 15 fishing vessels were authorized to fish for bigeye tuna in the Atlantic Ocean in 2006. Due to fishing pattern and the arrangement applied on fishing vessels in 2006 totally different from the years in the past, the study on the business running of fishing vessel is made on the basis of 220 tons bigeye catch quota allocated to individual vessel in 2006 comparing to the vessel in 2005. The research is conducted by using the method of Break-even analysis to analyze the cost-efficiency of vessels operation and by using the method of SWOT to analyze the impact on the fishing industry. From the results, it is found that the quota of bigeye tuna allocated to Taiwan is not commensurate with the fishing capacity of Taiwanese fishing fleet. The reasons for the adoption of the Resolution 05-02 by ICCAT are due mainly to part of Taiwan’s fishing vessels not complying with the international fisheries management measures, so as to be accused by Japan of engaging in IUU/FOC fish laundering. In 2006, the bigeye tuna quota allocation 4,600 tons are not enough to maintain normal operation of 72 ultra-low temperature tuna fishing vessels in the Atlantic Ocean. Among the 15 vessels targeting on bigeye tuna in the Atlantic, each vessel can share only 220 tons quota of bigeye tuna. However, the domestic administrative fishing regulations stipulated by the government, requiring that the fishing vessels enter port for inspection every three months and deploy observer on board each vessel, seemingly interrupted the fishing activities of the vessels. The actual fishing days and catch for each vessel on an average is 180 days and 163 tons respectively, a decrease of 73 fishing days comparing to the previous year (2005), while an increase of 163 tons of catch vice versa. As regards the impact on the fishing industry, the private sector and the government need to jointly assume the responsibility for the payment in dealing with vessel reduction and other related expenditure, thereby having direct impact on the employment which temporarily causes job loss of 3,000 workers, thus subsequently incurring the problem of bearing their families subsistence, as well as escalating the tension between Japan and Taiwan.