450 research outputs found
Ji yu shen du xie zuo xue xi ji shu de yi xue tu xiang yu yi fen ge
No full textPh.D.The computer-aided diagnosis in medical imaging computing has been a long-standing topic, in which the segmentation of abnormal regions plays an important role. Their accurate recognition and delineation can greatly assist the clinicians in early screening, grading, and monitoring of the disease. With the availability of the massive amount of data, deep learning has become a de facto standard approach in various medical image diagnosis applications. However, most deep learning methods adopt the single-task learning strategies, in which each task (e.g., classification, segmentation) is learned in isolation, thus these models are easily trapped in a single image feature space. To tackle this problem, we in this thesis present four novel collaborative learning strategies that exploit the more general and informative image representations across multiple correlated tasks (e.g., area and its contour) to address the abnormal region segmentation problems. The conducted comprehensive experiments demonstrate the effectiveness of our proposed collaborative methods.In the first part, we focus on the segmentation and delineation of colorectal polyps in colonoscopy video sequences. The proposed network is comprised of a shared encoder and two collaborative decoders targeted for polyp area and boundary segmentation respectively. Furthermore, a novel boundary-sensitive loss is proposed to model the interdependencies between the area and boundary branches, based on which the information of the two branches are reciprocally propagated and constrained, yielding a significant improvement in segmentation accuracy.For the second part, we study the segmentation task of four different types of fundus lesions. To exploit multi-scale image information, we propose a collaborative network architecture that comprises of a contextual branch and a local branch. An attention mechanism is designed to fuse feature maps from all decoding layers in order to effectively and fully combine informative features from the two branches. Moreover, an auxiliary classification task with a novel supervision scheme is introduced to jointly train the network, thus the problem of overfitting can be significantly reduced.In the third part, we investigate the tumor segmentation of histopathology images. In this work, we present a novel hybrid neural network for hepatocellular carcinoma (HCC) segmentation on H&E stained whole slide images (WSIs). Specifically, three task-specific branches, i.e., Brwhole_seg, Brtumor_seg and Brcls, are integrated to train the network, which enlarge the feature space, thus the network is able to learn more general features. To the best of knowledge, this is the first work on pixel-wise HCC segmentation with H&E stained WSIs.The training of deep learning based methods requires a large amount of annotated data, and the labeling task is an extremely time-consuming process for human annotators. Therefore, we in the fourth part propose a semi-automatic annotation method in order to reduce the annotation burden. The method is comprised of two subnetworks, a pixel-wise segmentation network and a polygon-based annotation network, which cooperate with each other. Our method integrates human annotators into the prediction loop, allowing to iteratively refine the predictions according to the suggestions from human annotators. We demonstrate the effectiveness of the human-network collaborative annotation, and achieve promising labeling results.在醫學影像計算領域中，電腦輔助診斷是⾧期存在的難題，其中醫療圖像病灶區域的分割尤為重要。病灶的準確識別和分割可以極⼤地幫助臨床醫⽣對疾病進⾏早期篩查，評估和治療。隨著海量資料的可⽤性，深度學習已經成為各類醫學圖像診斷應⽤中的標準⽅法。但是，⼤多數基於深度學習的⽅法都採⽤單任務學習策略，其中每個任務（例如分類，分割）都是獨⽴學習的，因此這些模型很容易陷⼊單個圖像特徵空間中。為了解決這個問題，我們在本⽂中提出了四種新穎的深度協作學習策略，這些策略利⽤跨多任務的更通⽤的圖像特徵，來處理病灶區域分割的任務。我們通過⼤量的實驗證明了我們提出的協作學習策略的有效性。在⽂章的第⼀部分，我們著重於結腸鏡視頻序列中⼤腸息⾁的分割，我們提出的網路由⼀個共⽤編碼器和兩個分別針對息⾁區域分割和邊界分割的解碼器組成。此外，我們提出了⼀種新穎的邊界敏感損失函數來類⽐區域分⽀和邊界分⽀的相互依賴性，基於此，這兩個分⽀的資訊可以相互傳播和約束，從⽽⼤⼤提⾼分割精度。在⽂章的第⼆部分，我們研究了四種不同類型的眼底病變的分割任務。為了利⽤多尺度圖像資訊，我們提出了⼀種協作網路，該網路結構包括上下⽂分⽀和局部分⽀。我們還設計了⼀種注意機制融合來⾃所有解碼層的特徵圖，以便網路可以有效地結合來⾃兩個分⽀的特徵。此外，我們還引⼊了⼀個新穎的分類任務分⽀來聯合訓練網路，從⽽⼤⼤地減少網路訓練過程的過度擬合問題。在⽂章的第三部分，我們研究了組織病理學圖像的分割任務。在這項⼯作中，我們提出了⼀種新穎的針對蘇⽊精-伊紅染⾊的肝細胞癌全玻⽚圖像的分割網路。具體來說，該網路集成了三個特定任務的分⽀，分別是Brwhole_seg，Brtumor_seg 和Brcls。三個分⽀的結合可以擴⼤網路的特徵空間，因此網路可以學習更通⽤的圖像特徵。據我們所知，這是第⼀個肝細胞癌全玻⽚圖像的分割⼯作。基於深度學習的網路訓練需要⼤量的標注資料，⽽標注⼯作極其耗時繁瑣。因此，我們在⽂章的第四部分提出了⼀種半⾃動標注⽅法，以減輕⼈⼯標注者的負擔。該⽅法由兩個⼦網路組成，分別是基於圖元的分割網路和基於多邊形的標注網路。我們的⽅法將⼈⼯標注者的建議集成到預測過程中，從⽽網路可以反覆運算地優化並完善標注結果。這裡我們通過實驗證明了⼈與網路協作標注⽅法的有效性。Fang, Yuqi.Thesis Ph.D. Chinese University of Hong Kong 2020.Includes bibliographical references (leaves 114-133).Abstracts also in Chinese.Title from PDF title page (viewed on March 30, 2022).Fang, Yuqi
Research on the Development of Voice Assistants in the Era of Artificial Intelligence
No full textVoice assistants have gradually occupied an important position in the products of many electronics companies. Artificial Intelligence voice assistants are able to interpret human speech and respond. Users can ask their assistant questions and manage other essential tasks such as email calendars through verbal commands. This paper analyzes the artificial intelligence voice assistant through the method of comparative analysis. The author studies the development situation of intelligent voice assistants, and compares the differences between Chinese and foreign voice assistants, and finally discusses the relationship between voice intelligent assistants and people’s lives. The author found that users in different countries have different functional preferences for using voice assistants, but they can help people’s work and life to a great extent. In other words, voice assistants play an important role in contemporary society. Therefore, people need to better understand the relationship between humans and machin
The lost tradition : changing interpretations of music in the three Chinese Confucian ritual classics from the Han to the Qing dynasty.
No full textSIGLEAvailable from British Library Document Supply Centre- DSC:DX185476 / BLDSC - British Library Document Supply CentreGBUnited Kingdo
Digital Integration of Medical Information (DIMI) in cardio-thoracic surgery in LUMC
No full textCommunication plays an essential role during open heart surgery. In order to improve the safety and the efficiency in the Coronary Artery Bypass Grafting (CABG) surgery. This research indicated three projects focusing on solutions of gathering and integrating medical data generated by the different machines and team members. Based on the observation of CABG surgeries, interviews with surgical members and meetings of new operating room design discussing criteria comparison, Project A introduced a new Operating Room design with a 75-inch nonmedical display showing medical data for better information sharing during surgeries. This solution was evidenced effective and will be implemented in the new operating room in practice. Project B simulated an auto-checklist of CABG surgery procedure to make sure every surgical team member aware of current situation during surgeries. Steps of surgeries and trigger of each step were determined with the assistance of cooperating surgical members. After negotiation and obtaining the consent of patients, medical information shown in EPD-Vision during their surgeries were allowed to be used. Then, necessary data from patients was obtained and processed by a MATLAB program. This simulation program uses the obtained data to detect automatically items of the checklist. As a result, times of the items of the checklist determined by the program were mostly comparable to the times written down by the assistant. In some cases, there was a time delay due to manual input by assistants. Hence, the project still has limitations to implement a completely automatic checklist.Project C is an APP design on a concept to remind related members of going back to the operating room quickly. Though it is on paper, most of surgical members showed positive attitude to this design.In conclusion, the integration of medical data from different medical instruments is a promising topic for further research
Identification of Neolithic Circular Enclosures through Aerial Imagery
No full textTraditionally, archaeological investigations, especially archaeological remains detection, mostly depend on human observation. In order to find the objects in large areas, a lot of fieldwork has to be done and it takes a long time for archaeologists to travel around. Nowadays, the development of LIDAR provides accurate 3D geometric information, which can be used for computer-based detailed terrain study. The application of deployment of computer vision methods also provides a new idea for the automatic object detection approach.In this study, the neural network architecture "ResNet18" was applied to airborne LiDAR data from the Western regions of Slovakia for the automated detection of undiscovered Neolithic Circular Enclosures (also called rondel in the thesis). NCEs are mysterious stone hedge-like rings scattered through Central/Eastern Europe. The LiDAR data was processed into digital rater data and realized data enhancement by the visualization technique -- Simple Local Relief Model (SLRM). Since the positive samples were limited, expanding the training dataset was crucial and was realized by data augmentation methods based on the positive samples of rondels. The augmented roundels were created by cropping the real roundels and pasting them on the new empty areas after slight modification. After that, the positive image samples and the same number of negative image samples constructed the whole data set and it was divided into two parts -- training data and test data. After the training process of ResNet18, the performances of deep learning models with different combinations of parameters were evaluated, and the selected model was applied to a large area (44276 × 29984 m2), the spatial distribution of the probabilities could be observed and 32 possible new rondel areas were chosen for further validation.Geoscience and Remote Sensin
Designing Bata Society: a Case Study of Zlín
No full textZlín, located in the east part of the Czech Republic, is a historical industrial town. The development of this modern city has a deep connection with Bata shoe company and its social scheme. The company has been known for its shoe production and the urban and architectural values from its pioneer and utopian town planning and its variant modules of building design. Inspired by English urban planner. Ebenezer Howard, Tomáš and several architects tried to achieve a modern living environment in Zlín by combining the aspects of country living with the conveniences of city life. One of Tomáš’ principles is “build collectively, live individually”; at a certain level, he realised his vision. Thousands of free-standing family housing were placed among the sizeable green land around the early 20s to late 30s for the Bata employees and their families. This thesis aims to emphasise the perspective of workers who work and live under the Bata system. The analysis of urban planning and architectural design in Bata’s factory town vividly reveals the life story of the working class on the land of Zlín, at the same time demonstrating how design as a medium helps to affect and shape the culture, society and even individual mentality.AR2A011Architecture, Urbanism and Building Science
From Border to Landscape: Designing a resilient landscape corridor in the Shenzhen-Hong Kong border area
No full textThe estuarine river landscape between Shenzhen and Hong Kong is a typical case where the landscape lost its value and characteristics due to the border policy. The two megacities Shenzhen (a city in mainland China) and Hong Kong (a special administrative region), are located in the southeast of the Pearl River Delta, one of the world's largest and rapidly developing delta regions.The Shenzhen River in the middle of the two cities has been a border river since 1898. Coupled with the green buffer zone on the HK side that was once a restricted border area, the bay, the estuary, the river, and the land together form a uniquely polarized landscape, with urbanization and nature across the river. Whether to protect the natural environment or develop the land; and whether to preserve the ecological value or restore the aquaculture productivity become urgent arguments regarding future development in this border area.This thesis aims to explore the development of a resilient landscape corridor across the Shen-Kong border to achieve the co-development of the environment and society. Viewing the landscape as a multifunctional and operative field with its own spatial, ecological and socio-cultural qualities (Nijhuis & Jauslin, 2015) can be a way to respond to the aforementioned phenomenon. The thesis focuses on applying the theory of landscape infrastructure as a method in research and design to generate a resilient landscape framework that is considered armature for urban and rural development to facilitate interactions between natural and human systems (Nijhuis & Jauslin, 2015), projecting natural processes and performing multiple functions. Moreover, a methodology for designing towards a landscape framework is given in this thesis. Examples are provided to show the application of landscape infrastructure services and the assessment.Architecture, Urbanism and Building Sciences | Landscape Architectur
- …
