DSpace collection: 研究計畫

三維立體資料之多方向投影語義分割法;Multi-Directional Projection Semantic Segmentation for Volumetric Data

Mon, 13 Jan 2020 06:49:21 GMT

title: 三維立體資料之多方向投影語義分割法;Multi-Directional Projection Semantic Segmentation for Volumetric Data abstract: 在許多預防性的疾病篩檢項目中，高解析度三維立體醫學影像資料(例如:核磁共振造影、細切的電腦斷層掃描)是目前最精確的非侵入性之病灶偵測和惡性程度分析方法。但是從高解析度三維立體資料偵測細微病灶，目前仍然是一件特別困難的差事，主要原因是，在二維影像中，細微的病灶和許多身體細結構相類似而且非常容易混淆。雖然，放射專科醫師可以從連續幾張二維圖像在腦中重建三維資料做區分，一般深度學習的卷積神經網路還是大部分只能處理二維影像。所以在此研究計畫書，我們提出一個從三維立體影像匯集多個不同方向的二維正射投影圖之總體，經過目前最好的語義分割卷積神經網路辨識後，這些各種投影方向的二維分割資料可以被反推回原來的三維位置去累積病灶位置的可能性高低地圖。從我們的一個先期性研究中所得到的結果分析，從多個不同方向的二維正射投影圖之分割資料總體回推三維形狀結構資料是準確率高而且有臨床應用價值的新研究方法和方向。 ;High spatial resolution volumetric data such as MRI and spiral CT scanning are state-of-the-art, non-invasive screening methods for the detection of lesions and the analysis of malignancy. However, detecting small lesions from volumetric data remains difficult because many anatomic variations resemble true lesions in 2D images. While radiologists can browse through consecutive 2D images to visualize the 3D structure in mind, existing deep learning convolutional neural networks (CNNs) are mostly designed for 2D images. In this research proposal, we envision an ensemble of multi-directional 2D orthographic projection views (by volume rendering) of a subvolume of interest such that a partially obscured lesion can at least be detected from some directions by existing semantic segmentation CNNs. These 2D segmentation data are then reversely projected back to a 3D-probability-accumulation map to identify lesions. Experimental results of our pilot study indicate the effectiveness of building 3D context information from multi-directional 2D projection views.

EMD的局部性與模態混之關係--數學理論建立以及它的應用;The Relation between Locality and Mode-Mixing of the Empirical Mode Decomposition –The Theoretical Foundation and Its Applications

Mon, 13 Jan 2020 06:49:02 GMT

title: EMD的局部性與模態混之關係--數學理論建立以及它的應用;The Relation between Locality and Mode-Mixing of the Empirical Mode Decomposition –The Theoretical Foundation and Its Applications abstract: 經驗模態分解(EMD)方法是一個非線性、自適應的時頻分析方法。它藉由幾次的篩選算子作用後將一個複雜時間訊號拆解為若干個本質模態函數(intrinsic mode function、IMF)；這些振盪的IMF往往對應到某些物理或生理現象。由於EMD是建立在極值點所暗示的局部尺度為出發點，因此EMD一直被假設是一個局部性的方法。局部性是指訊號任一個時間點的動態特性經過篩選算子作用後，只會影響其附近的點。而另一方面當極值點分佈不均勻時EMD又會產生所謂的模態混合。它是指同一個IMF中同時摻雜著非穩態的高頻與低頻訊號而模糊了IMF的時頻意義，但如今尚無定量分析；我們將根據極值點分佈距離定義局部不均勻度(DONU)。我們將運用我們過去所證明之篩選算子脈衝響應定理著手去證明對一次篩選算子而言，當DONU超過特定上閥值時，EMD將喪失居部性，而低於特定下閥值時，將能保證其局部性。本研究將找出這上下閥值，並證明其EMD 存在與喪失居部性的關聯。我們也將推論並驗證模態混和發生原因及其與EMD對該訊號喪失局部性的關係，進一步推論與證明脈衝響應定理對多次篩選是否成立，並探討多次篩選算子作用後的局部性。除了數學理論建立，基於本研究所獲得的結論，我們將提出一個EMD改善方法，使其未來應用更具理論與學理基礎。 ;Empirical Mode Decomposition (EMD) is a nonlinear, adaptive time frequency method. A signal is decomposed into a series of intrinsic mode functions (IMFs) by an iterative sifting process. Each IMF often corresponds to a particular physical oscillation. Since EMD is based on the local scale of, it has been hypothesized that EMD is a local method. Locality means that a time point would mainly affect its neighboring points after applying the sifting iterations. On the other hand, EMD would suffer from the mode-mixing phenomenon when the extrema distribution is highly non-uniformly distributed, but there is no a clear quantatively definition till now. We will define the first order of non-uniformity (DFONU) based on the extrema distribution. We will apply the impulse response theorem of the sifting operator to prove that EMD is only a conditionally local method. Yo be more specific, as DFONU is greater then 3.19, EMD will lose its locality, as DFONU is less than 2.45, EMD will be local. We will then explain that the reason for causing mode-mixing is when EMD lose its locality. We will then prove that the impulse response theorem can be extended from single sifting iteration to multiple iterations and study how the locality will be affected by continuous siftings. Finally we will propose a locality-based improved EMD algorithm based on these new finding.

複雜系統實驗資料的適應性數據分析方法;Adaptive Data Analysis Approaches for Empirical Data from Complex Systems

Thu, 20 Dec 2018 05:50:15 GMT

title: 複雜系統實驗資料的適應性數據分析方法;Adaptive Data Analysis Approaches for Empirical Data from Complex Systems abstract: 本計畫係為期三年（2018年8月至2021年7月）的基礎研究計畫，研究主題為「複雜系統實驗資料的適應性數據分析方法」，主要目的係以適應性數據分析方法研究複雜系統的實驗資料，從中理出系統的本徵特性，進而建構對應的物理模型。本計畫擬探討的具體問題包括：1. 基於量測量之最大資訊量的適應性數據分析架構2. 蛋白質的力致活化動力學3. 實驗生醫信號分析4. 對應於金融系統數據特性的理論力學及臨界性質本計畫可望達成的目標有：1. 發展出適於複雜系統研究之基於最大資訊量之量測量的適應性數據分析方法2. 建立以力致活化的奈米力學識別蛋白質功能局域的一般性理論，並應用於分子對接問題3. 將所建構之適應性數據分析方法應用於實驗生醫信號分析4. 了解金融數據的性質、改良現有以經驗法則為基礎的模式，及建立對應的力學理論本計畫的研究成果對相關領域的基礎研究具有應用與參考價值。本計畫將有兩位博士生參與上述具體問題；計畫執行期間可望每年在SCI期刊發表數篇研究論文。 ;This is a three-year-term (2018.8~2021.7) research project entitled “adaptive data analysis approaches for empirical data from complex systems”. The aim of the project is to study empirical data from complex systems using adaptive data analysis approaches, and from which to extract intrinsic properties of the systems, thereby to construct corresponding physics models. Specifically, this project plans to deal with the following problems: 1. Adaptive data analysis frameworks based on the maximum information content of measures2. Dynamics of protein mechanoactivation3. Empirical biomedical signal analysis4. Theoretical mechanics corresponding to the properties of financial data and its critical propertiesThis project will reach the following achievements:1. Adaptive data analysis approaches based on the maximum information content of measures for complex systems2. A general theory for identifying protein functional domains from nanomechanics of mechanoactivation, and its applications to molecular docking problems3. Applications of the developed adaptive data analysis approaches on empirical biomedical signals4. Better understanding of the properties of financial data, improving the experience-based models, and developing its mechanical correspondenceOur research results will provide useful references for fundamental researches of related fields. Two PhD students will involve in this project and join the study of the above-mentioned research problems. A number of research papers can be published in SCI Journals every year, based on the results of this project.

改進經驗模態分解法擷取間歇性生醫訊號的正確性;Improving Empirical Mode Decomposition Accuracy in Extracting Intermittent Biomedical Signals

Wed, 02 Aug 2017 04:06:05 GMT

title: 改進經驗模態分解法擷取間歇性生醫訊號的正確性;Improving Empirical Mode Decomposition Accuracy in Extracting Intermittent Biomedical Signals abstract: 研究期間：10508 ~ 10607