http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/91676 Sat, 04 Apr 2026 04:50:07 GMT 2026-04-04T04:50:07Z http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/93775 標題: 透過可解釋的認知評估，研究轉移、少樣本和元學習在基於靜息態和聽覺穩定狀態反應腦電圖的神經退化性疾病診斷之生物標記識別; Investigating Transfer, Few-Shot, and Meta Learning for Biomarker Identification in Resting-State and Auditory Steady-State Response EEG-Based Diagnosis of Neurodegenerative Diseases with Interpretable Cognitive Assessments 作者: 楊美美; DANIELLE PENELLA P. YU 摘要: 失智症患者的靜息態和聽覺穩定狀態反應 (ASSR) 相關腦電圖 (EEG) 訊號，如腦電圖減慢、伽瑪相位鎖定和功率調變改變，通常能揭露患者的認知異常。傳統上，腦電圖的分析涉及使用功率譜密度(PSD) 地形圖、試驗間一致性(ITC) 計算和事件相關頻譜擾動(ERSP) 分析等技術將訊號轉換為空間頻率和時間頻率之影像表示。這樣的方法可以凸顯關鍵的細微差別以進行準確的診斷和解釋。隨著深度學習 (DL) 在推進診斷的關鍵解決方案展露頭角，透過可解釋的人工智慧 (xAI) 進行，它為傳統方法提供了一種可行、高效且重要的替代方案。而深度學習也揭示了腦電圖作為篩檢工具並協助找出神經退化性疾病個體生物標記的能力。 在這項研究中，我們專注於整合遷移學習、小樣本學習 (FSL) 和元學習技術，以最大化深度學習的潛力。我們收集了阿茲海默症 (AD) 、罕見疾病額顳葉失智症 (FTD) 和精神分裂症 (SZ)患者的腦電圖數據，以研究與失智症相關的差異特徵。閉眼時的靜息態腦電圖資料收集自 15 位AD患者、15 位 FTD 患者、16位SZ 患者以及 15位健康對照組 (HC)。此外，針對 ASSR 數據集我們收集了以40 Hz 刺激的數據，其中包括 10 名SZ 患者和 15 名 HC。 我們的方法採用預訓練的捲積神經網路 (CNN) 以及 miniImageNet 資料集來有效地分析不同腦電圖資料集。我們結合了FSL 和元學習，以增強模型在2-way 5-shot分類中的泛用性和適應性，其中每個episode 涉及從兩個類別中隨機採樣五個樣例，並在100 多個episode的穩健測試中對三個查詢進行測試。儘管神經退化性疾病研究中的數據可用性是個重大挑戰，而且我們的數據集規模相對較小，但相似的人類上皮細胞 2 型 (HEp-2 Cell) 數據集和血細胞計數和檢測 (BCCD）白血球數據集達到顯著的進步。 其中，HEp-2 細胞資料集包含13,596 張影像，分為六類（著絲粒、高爾基體、均質、核仁、核膜和斑點型），而BCCD 白血球資料集包括五類（嗜中性球、淋巴細胞、單核細胞、嗜酸性粒細胞和嗜鹼性粒細胞），總共12,436張影像，達到了55.1%的準確率。 相對 α 波段功率是靜息態腦電圖最顯著的特徵。使用含有殘差網路18 (ResNet-18) 的原型網路(prototypical network)和帶有ResNet-34 編碼器的匹配網路(matching network)，我們分別實現了82.83% 和76.33% 的平均準確度。其中最具辨別能力的類別對是AD 和SZ（94.44%和 92.22% 的準確率）。相較之下，我們的結果表明，P4 和 Fz 通道在 ASSR 間相干性方面實現了最高性能，平均準確度分別為 68.83% 和 70.00%。另一方面，與事件相關的光譜擾動發現T5 和 Fz 通道是有區別性的，兩個模型的平均表現為 71.17% 和 69.33%。 為了更深入研究神經動力學，我們實作了梯度加權類活化映射 (Grad-CAM)，這是 xAI 中的一個突出方法。這種方法大大增強了我們對疾病狀態和病理機制中振盪行為的理解，並可能有助於生物標記的辨識。 Grad-CAM 顯示頂葉的相對 α 功率降低，以及與靜止狀態下認知障礙的潛在關聯性。此外，基於ASSR 的Grad-CAM視覺化揭示了誘發功率和鎖相中明顯的伽馬帶振盪，表明N-甲基-D-天冬氨酸受體(NMDAR) 拮抗劑可能引起伽馬氨基丁酸(GABA) 中間神經元功能障礙。證據顯示，在 ASSR 任務期間，Fz 通道 ITC和ERSP在Grad-CAM 上減少，顯示 SZ 和 HC 受試者存在異常的神經同步和頻譜動態模式。這些神經損傷的模式會影響神經網路的完整性，而能揭示特定的大腦區域如何導致疾病。 總之，結合傳統腦電圖分析方法和新穎的深度學習技術使我們能夠解決神經退化性疾病研究中的關鍵挑戰。我們的方法不僅提高了其穩健性和有效性，還提高了複雜神經數據的可解釋性。因此，我們的方法能啟發新的臨床方案發展，以及加速新藥標的的發現。; Resting-state and auditory steady-state response (ASSR)-related electroencephalography (EEG) signals in individuals with dementia commonly reveal cognitive abnormalities, such as EEG slowing, gamma phase locking, and power modulation alterations. Traditionally, EEG analysis involves transforming signals into spatial frequency and time-frequency image representations using techniques via power spectral density (PSD) topographic maps, intertrial coherence (ITC) calculations, and event-related spectral perturbation (ERSP) analysis, highlighting crucial nuances essential for accurate diagnosis and interpretation. With the emergence of deep learning (DL) as a pivotal solution in advancing diagnostics, it offers a feasible, efficient, and crucial alternative to conventional methods for accurate differentiation, complemented by interpretation through explainable artificial intelligence (xAI). DL methods, in particular, reveal the ability of EEG to serve as a screening tool and assist in identifying biomarkers in individuals with neurodegenerative diseases. In this study, we focus on integrating transfer learning, few-shot learning (FSL), and meta-learning techniques to maximize the potential of DL. EEG data from individuals with conditions such as Alzheimer’s disease (AD), a rare disease frontotemporal dementia (FTD), and schizophrenia (SZ) were collected to investigate differential features associated with dementia. Resting-state EEG data from the eyes-closed protocol were collected from 15 patients diagnosed with AD, 15 with FTD, 16 with SZ, as well as from 15 healthy controls (HCs). In addition, 40-Hz stimulus data for the ASSR dataset, consisting of 10 individuals with SZ and 15 HCs, were obtained. Our approach adopted pretrained convolutional neural networks (CNNs) along with the miniImageNet dataset benchmark to analyze diverse EEG datasets effectively. We incorporated FSL and meta-learning to enhance the models’ generalizability and rapid adaptability in 2-way 5-shot classification, where each episode involved randomly sampling five support examples from two classes and robust testing on three queries over 100 episodes. Despite major challenges with data availability in neurodegenerative disease research and the relatively small size of our datasets, significant improvements have been observed in similar test scenarios involving the Human Epithelial Cells type 2 (HEp-2 Cell) dataset and the Blood Cell Count and Detection (BCCD) White Blood Cell dataset. The HEp-2 Cell dataset comprises 13,596 images divided into six classes (Centromere, Golgi, Homogeneous, Nucleolar, Nuclear Membrane, and Speckled). In contrast, the BCCD White Blood Cell dataset includes five classes (Neutrophil, Lymphocyte, Monocyte, Eosinophil, and Basophil), totaling 12,436 images, achieving its highest accuracy of 55.1%. Relative alpha band power is the most distinguishing feature of resting-state EEG. Using a prototypical network with residual network-18 (ResNet-18) and a matching network with ResNet-34 encoders, we achieved mean accuracies of 82.83% and 76.33%, respectively, with the most discriminative class pair being AD and SZ (94.44% and 92.22% accuracies). In contrast, our results indicate that the P4 and Fz channels achieve the highest performance in ASSR intertrial coherence, with average accuracies of 68.83% and 70.00%, respectively. On the other hand, event-related spectral perturbations identify the T5 and Fz channels as discriminative, averaging 71.17% and 69.33% for both models. To delve deeper into neural dynamics, we implemented gradient-weighted class activation mapping (Grad-CAM), a prominent feature attribution in xAI. This method has greatly enhanced our understanding of oscillatory behavior in disease states and pathological mechanisms, potentially aiding in biomarker identification. The Grad-CAMs showed decreased relative alpha power in the parietal lobes and its possible association with cognitive impairment in the resting state. Moreover, ASSR-based Grad-CAM visualizations revealed distinct gamma band oscillations in evoked power and phase locking, suggesting probable gamma-aminobutyric acid (GABA)ergic interneuron dysfunction caused by N-methyl-D-aspartate receptor (NMDAR) antagonists. The evidence showed a reduction in Fz channel ITC and ERSP Grad-CAMs during the ASSR task, indicating aberrant neural synchrony and spectral dynamics patterns across SZ and HC subjects. These patterns underlying neurological impairments affect neural network integrity, thereby revealing how specific brain regions contribute to diseases. In conclusion, combining traditional EEG analysis methods and innovative DL techniques has allowed us to address critical challenges in neurodegenerative disease research. Our methodology improves its robustness and effectiveness and the interpretability of complex neural data. Thus, our approach may inspire the development of new clinical practices and expedite the discovery of new drug targets. Mon, 01 Jan 2024 00:00:00 GMT http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/93775 2024-01-01T00:00:00Z http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/94926 標題: 注意力深度學習方法應用於時間序列腦電波圖針對心跳停止後腦神經損傷的預後預測; An Attention-Based Deep Learning Approach of Using Time-Series EEG for Predicting Neurological Outcomes in Cardiac Arrest 作者: 曾世傑; Jefferson Sy Dionisio 摘要: 突發性心臟驟停（SCA）患者通常因缺氧時間過長而陷入昏迷，醫師需提供神經系統預後，協助臨床決策。本研究旨在利用早期腦電圖（EEG）數據訓練Transformer模型，預測SCA昏迷患者的神經系統預後。Transformer模型利用自注意力機制從長序列中學習模式。我們利用完整的小時級EEG序列，將其分割為5分鐘的時段，使模型能夠捕捉長距離的時間序列模式。通過將每個EEG序列視為訓練樣本，我們增加了數據樣本量，提高了模型學習特定記錄模式的能力。預測結果按患者進行了整合評估。專注於EEG數據的模型展現出了良好的預測性能，在保留測試集上的AUROC為0.82，AUPRC為0.90，在外部測試集上的AUROC為0.73，AUPRC為0.93。本研究凸顯了注意力機制在識別EEG序列中時間模式方面的潛力，提升了對SCA患者預後的能力。; Surviving sudden cardiac arrest (SCA) patients often remain in a coma due to a prolonged lack of oxygen, requiring physicians to provide prognoses on neurological outcomes to aid in clinical decisions. This study aims to predict neurological outcomes in SCA coma patients using early electroencephalogram (EEG) data to train a Transformer model, which leverages self-attention to learn patterns from lengthy sequences. We utilized full hours of EEG sequences, subdividing them into 5-minute epochs, allowing the model to capture long-distance time series patterns. By treating each individual EEG sequence as a training sample, we increased our data sample size and improved the model's ability to learn recording-specific patterns. Predictions were aggregated for patient-wise evaluation. Focusing exclusively on EEG data, our model demonstrated promising predictive performance, with an AUROC of 0.82 and an AUPRC of 0.90 on the holdout test set, and an AUROC of 0.73 and an AUPRC of 0.93 on an external test set. This study underscores the potential of attention mechanisms to discern temporal patterns in EEG sequences, enhancing SCA patient prognosis. Mon, 01 Jan 2024 00:00:00 GMT http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/94926 2024-01-01T00:00:00Z http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/101778 標題: 機器學習與深度學習於心臟病發作預測之比較分析; Comparative Analysis of Machine Learning and Deep Learning Approaches for Heart Attack Prediction 作者: 莫子佩; Muhammad Aulia Rahman 摘要: 心血管疾病仍然是全球主要的死亡原因，其中心肌梗塞（心臟病發作）因其突發性及可能致命的特性，對臨床及公共衛生構成特別嚴重的挑戰。因此，準確且及時地預測心肌梗塞風險對於有效的預防與早期介入至關重要。近年來，機器學習（ML）與深度學習（DL）技術已逐漸被應用於心血管風險預測；然而，針對其相對效能，尤其是在結構化臨床資料上的應用，目前仍缺乏共識。 本研究針對心肌梗塞預測，對多種機器學習及深度學習模型進行系統性的比較分析，所使用的結構化臨床資料集包含918位病患的基本人口學資料、臨床檢查資訊及生活型態特徵。傳統的機器學習模型包括羅吉斯迴歸（Logistic Regression）、支持向量機（Support Vector Machines）、隨機森林（Random Forest）、梯度提升（Gradient Boosting）及XGBoost，與深度學習模型，包括多層感知器（MLP）、基於Keras的深度神經網路（DNN）以及TabNet，進行比較。所有模型皆在相同實驗條件下訓練與評估，採用統一的資料前處理流程、分層的訓練–測試集劃分，以及標準化的評估指標。模型效能評估指標包括準確率（Accuracy）、精確率（Precision）、召回率（Recall）、F1分數以及受試者工作特徵曲線下面積（ROC-AUC），其中特別強調召回率及ROC-AUC，因其在臨床上對於降低漏診具有高度相關性。 實驗結果顯示，傳統機器學習模型，特別是集成方法，在此結構化資料集上的表現普遍優於深度學習模型。隨機森林取得最高的整體準確率（90.22%）與F1分數（91.35%），同時維持高召回率（93.14%）及具有競爭力的ROC-AUC（0.933），顯示其能有效辨識高風險病患。使用RBF核的支持向量機則取得最高ROC-AUC（0.949），反映出其在辨別高風險與非高風險病患之間的能力最佳。相比之下，深度學習模型表現雖然具有競爭力，但略低於傳統機器學習，其中Keras DNN達到89.67%的準確率與0.949 ROC-AUC，MLP稍低，而TabNet表現較弱（82.61%準確率、0.912 ROC-AUC），可能受限於資料量較小以及特徵豐富度不足。 這些結果顯示，對於中等規模的結構化臨床資料，集成機器學習模型比深度學習模型更有效且實用。除了預測效能之外，像羅吉斯迴歸與隨機森林這類具可解釋性的模型，能讓臨床人員理解各特徵對預測結果的貢獻，對臨床採用至關重要。總體而言，本研究強調模型選擇應與資料特性及臨床需求相匹配，提供實證基礎的建議，以協助開發可靠的心肌梗塞預測系統，應用於醫療決策支援。; Cardiovascular diseases remain a leading cause of global mortality, with heart attacks presenting a particularly severe clinical and public health challenge due to their sudden onset and potentially fatal outcomes. Accurate and timely prediction of heart attack risk is therefore essential for effective prevention and early intervention. In recent years, both machine learning (ML) and deep learning (DL) techniques have been increasingly applied to cardiovascular risk prediction; however, there remains limited consensus regarding their relative effectiveness, particularly when applied to structured clinical datasets. This study presents a systematic comparative analysis of selected machine learning and deep learning models for heart attack prediction using a structured clinical dataset comprising 918 patient records with demographic, clinical, and lifestyle-related features. Traditional ML models including Logistic Regression, Support Vector Machines, Random Forest, Gradient Boosting, and XGBoost were compared with DL approaches, namely a multilayer perceptron (MLP), a Keras-based deep neural network (DNN), and TabNet. All models were trained and evaluated under identical experimental conditions, employing a unified preprocessing pipeline, stratified train–test split, and standardized evaluation metrics. Model performance was assessed using accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (ROC-AUC), with particular emphasis on recall and ROC-AUC given their clinical relevance in minimizing missed diagnoses. The experimental results demonstrate that traditional machine learning models, particularly ensemble-based methods, generally outperformed deep learning approaches on this structured dataset. Random Forest achieved the highest overall accuracy (90.22%) and F1-score (91.35%), while maintaining strong recall (93.14%) and a competitive ROC-AUC (0.933), demonstrating its ability to correctly identify at-risk patients. SVM with an RBF kernel attained the highest ROC-AUC (0.949), reflecting superior discriminative capability between patients at risk and those not at risk. In contrast, deep learning models exhibited comparable but slightly lower performance, with the Keras DNN achieving 89.67% accuracy and 0.949 ROC-AUC, the MLP slightly lower, and TabNet performing substantially weaker (82.61% accuracy, 0.912 ROC-AUC), likely due to the relatively small dataset size and limited feature richness. These findings suggest that, for structured tabular clinical data of moderate size, ensemble machine learning models offer a more effective and practical solution than deep learning approaches. Beyond predictive performance, interpretable models such as Logistic Regression and Random Forest allow clinicians to understand how individual features contribute to predictions, which is crucial for clinical adoption. Overall, this study underscores the importance of aligning model selection with both data characteristics and clinical requirements, providing evidence-based guidance for the development of reliable heart attack prediction systems in healthcare decision support contexts. Thu, 01 Jan 2026 00:00:00 GMT http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/101778 2026-01-01T00:00:00Z http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/97541 標題: 台灣族群膽結石疾病遺傳易感性研究： 基因-表型組學方法與多基因風險評分模型; Genetic Susceptibility of Gallstone Disease in Taiwanese Population: A Genome-Phenome Approach with Polygenic Risk Score 作者: 蔡岳君; Yueh-Chun Tsai 摘要: 膽結石是種常見消化系統疾病，復發率與衍生相關疾病造成醫療系統極大負擔。膽結石形成與膽汁酸代謝、膽固醇代謝失衡、基因背景與種族差異密切相關，但大多數膽結石疾病基因研究主採用西方人群數據庫，限制對東方人群治病性生理機轉的理解。因此，本研究利用台灣人體生物資料庫，經品質控制後共108,403名具漢族受試者進行全基因體關聯分析（GWAS），探討與膽結石相關的遺傳變異，亦進行 LDSC 分析驗證通膨程度（λGC = 1.0741；intercept = 1.0067），結果顯示通膨現象主要源自多基因性而非其他混雜因子。建構多基因風險預測模型（Polygenic Risk Score, PRS），PRS-CS 展現極高預測力（AUC = 0.98）。本研究鑑定出57個顯著基因變異位點，其中有38 個新發現，計算每個顯著基因變異位點的治病機率，以 rs80217587（TM4SF4）具有最高的後驗包含機率（PIP = 0.202），顯示其為最具潛力的致病變異位點。本研究發現的7個Lead SNP，分別對應1個控制區（RP11-626H12.1）與6個鄰近基因（TM4SF4、LRBA、UBXN2B、CYP7A1、HNF4A 和 ANO1），其中TM4SF4、LRBA、UBXN2B 和 ANO1 為首次在東亞族群提出與膽結石易染性的關聯基因。我們的研究結果指出 TM4SF4 可作為膽結石疾病早期偵測與預防性醫療管理的潛在生物標記與治療標的。。透過路徑富集分析，我們也提出三組血中生物標記群（GGT、ALT、CRP、膽酸/膽紅素前驅物等），可望作為早期非侵入性風險預測工具。搭配多基因風險分數與個人飲食改善計畫建議，有助於建立針對膽結石高風險族群之精準預防策略。; Gallstone disease (GSD) is a prevalent gastrointestinal disorder with high recurrence and substantial healthcare burden. Its etiology involves bile acid and cholesterol metabolism, genetic predisposition, and ethnic variation. However, most genetic studies focused on Western populations, limiting insights into East Asian-specific mechanisms. This study leveraged data from 108,403 Han Chinese individuals in the Taiwan Biobank to perform a genome-wide association study (GWAS) on GSD. After quality control and adjustment for population stratification, linkage disequilibrium score regression (LDSC) confirmed minimal inflation (λGC = 1.0741; intercept = 1.0067), suggesting that observed signals are due to polygenicity. Polygenic risk prediction using PRS-CS demonstrated strong performance (AUC = 0.98). We identified 57 significant SNPs, including 38 novel variants. Bayesian fine-mapping revealed rs80217587 (TM4SF4) as the top causal candidate (PIP = 0.202), along with 7 lead SNPs mapped to TM4SF4, LRBA, UBXN2B, CYP7A1, HNF4A, ANO1, and a non-coding region (RP11-626H12.1). TM4SF4, LRBA, UBXN2B and ANO1 are first mentioned in East Asian population. We highlight TM4SF4 as a potential biomarker and therapeutic target for the early detection and preventive management of GSD. Moreover, pathway enrichment analysis identified candidate biomarkers (GGT, ALT, CRP, bile acid precursors) for potential early, non-invasive risk laboratory test screening. By integrating PRS with personalized dietary intervention plans, we propose a precision prevention strategy tailored for individuals at high risk of GSD. Wed, 01 Jan 2025 00:00:00 GMT http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/97541 2025-01-01T00:00:00Z