http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/102760 Program for Precision Health and Intelligent Medicine 2026-08-21T01:20:18Z http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/102761 標題: 開發以密度為基礎的空間功能分析框架： 解析 HER2 陽性乳癌中的功能區域與腫瘤休眠異質性; Development of a Density-Based Spatial Functional Analysis Framework for Characterizing Tumor Functional Organization and Dormancy Heterogeneity in HER2-Positive Breast Cancer 作者: 劉凡箴; Fan- Jhen Liu 摘要: 空間轉錄體技術可在保留組織結構下分析基因表現，有助於探討細胞空間分布與腫瘤微環境。然而如何解析空間中生物功能狀態分布仍具挑戰性。現有分析方法多延用自單細胞RNA定序流程，主要著重於單一基因表現或空間變異基因，即使結合 GSEA 或 GSVA 等路徑分析，在面臨基因表現訊號微弱或分布不均時，仍難以完整捕捉真實生物功能狀態的連續性空間變化，也缺乏在組織空間中明確辨識功能富集區域的能力。為克服上述限制，本研究提出一套以密度為基礎之分析框架 KDESH（KDE-based Spatial Hotspot Detection），用以辨識具有空間聚集特徵之生物功能路徑。此方法結合路徑活性評分、核密度估計（kernel density estimation）及置換檢定（permutation-based statistical testing），建立具統計基礎的空間功能分布模型。相較於傳統方法，KDESH不需依賴預先劃分之空間區域，而能捕捉功能途徑在空間上的連續性變化，並進一步辨識局部功能活性顯著富集的空間熱點（spatial hotspots）。此外，本方法亦可量化不同生物途徑之間的空間共定位（co-localization）。將KDESH應用於HER2 陽性乳癌空間轉錄體資料，並以腫瘤休眠作為具體的生物學探討案例。分析發現多數生物功能途徑在腫瘤組織中呈現顯著空間聚集現象，是腫瘤微環境的一項普遍特徵。另外，具有休眠特徵之腫瘤上皮細胞亦呈現顯著空間聚集 (p < 0.05)，進一步分析發現，腫瘤休眠並非單一的細胞狀態，而是呈現空間與功能異質性。根據所處微環境條件，休眠細胞可區分為缺氧相關休眠熱點（hypoxia-associated dormancy hotspots, HDH）與常氧相關休眠熱點（normoxia-associated dormancy hotspots, NDH），代表不同的功能狀態與潛在調控機制。本研究建立的 KDESH 框架，可系統性解析腫瘤組織中的功能空間分布，並量化不同生物路徑之間的空間關聯，有助於對腫瘤微環境空間結構的理解，亦突顯微環境所驅動之功能異質性在癌症生物學中的重要性，並可作為未來空間腫瘤生物學研究與微環境導向治療策略發展之基礎。; Spatial transcriptomics can analyze gene expression while preserving tissue structure, which helps in exploring the spatial distribution of cells and the tumor microenvironment. However, analyzing the spatial distribution of biological functional states remains challenging. Current analytical methods mostly follow the single-cell RNA sequencing pipeline, mainly focusing on single-gene expression or spatially variable genes. Even when combined with pathway analysis such as GSEA or GSVA, they still struggle to fully capture the continuous spatial changes in true biological functional states when faced with weak or unevenly distributed gene expression signals, and cannot clearly identify functionally enriched regions in tissue. To address these limitations, we introduce a density-based analytical framework, KDESH. This framework combines pathway activity scoring, kernel density estimation, and permutation-based statistical testing to establish a statistically based spatial functional distribution model. Compared to traditional approaches, KDESH does not depend on predefined spatial regions but can capture continuous spatial changes in functional pathways and further identify spatial hotspots where local functional activity is significantly enriched. Furthermore, this framework can also quantify the spatial co-localization between different biological pathways. We applied KDESH to HER2-positive breast cancer spatial transcriptome data, using tumor dormancy as a case study. This analysis revealed that most biological functional pathways exhibit significant spatial aggregation in tumor tissues, indicating that such spatial functional organization is a common feature of the tumor microenvironment. Additionally, dormancy-like cells showed significant spatial aggregation (p < 0.05), indicating that these cells tend to accumulate in specific microenvironmental niches. Further analysis suggested that tumor dormancy is not a single cellular state, but rather exhibits spatial and functional heterogeneity. Based on their microenvironment, they can be classified into hypoxia-associated dormancy hotspots (HDH) and normoxia-associated dormancy hotspots (NDH), representing different functional states and potential regulatory mechanisms. Overall, the KDESH framework provides a novel approach to explore the spatial functional distribution within tumor tissues and to quantify the spatial relationships between different biological pathways. This deepens our understanding of the spatial architecture of the tumor microenvironment, highlights the importance of microenvironment-driven functional heterogeneity in cancer biology, and can be used as a foundation for future spatial tumor biology research and the development of microenvironment- guided therapy strategies. 2026-01-01T00:00:00Z http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/102881 標題: 自律神經系統相關參數與焦慮及憂鬱量表所得分數之關聯性研究; A Study of Association between Autonomic Nervous System-Related Parameters and Scores Obtained from Anxiety and Depression Scales 作者: 吳旻璇; Min-Hsuan Wu 摘要: 焦慮與憂鬱症狀與自律神經系統失調之間的關聯受到廣泛關注，而心律變異度 (Heart Rate Variability, HRV) 被視為評估自律神經調節的重要非侵入性生理指標。然而，過去多數研究主要著重於靜態檢測，較少探討生理挑戰條件下的動態自律神經反應。因此，本研究使用氣泵式行動心律變異血壓檢測儀，探討不同姿勢條件下自律神經指標與焦慮及憂鬱量表分數之間的關聯性，並評估姿勢挑戰是否能提供較靜態測量更具生理意義之資訊。本研究共納入 90 位受試者，使用 GAD-7、ASI-3、PHQ-4 與 BDI-II 評估焦及憂鬱症狀，並於坐姿與站姿條件下量測 HRV、低頻功率 (LF)、高頻功率 (HF) 及 LF/HF ratio。研究結果顯示，整體 HRV 由坐姿條件下的 40.41 ± 4.34 顯著下降至站姿條件下的 38.82 ± 6.33 (p = 0.014)，顯示姿勢轉換成功誘發可測量之自律神經調節反應。雖然 HRV 與心理量表之相關性整體偏弱，但站姿條件下所呈現之反應模式與焦慮及憂鬱相關之自律神經失調理論較為一致。較高之心理症狀分數傾向與較低之 HRV 與 HF，以及較高之 LF 指標有關，反映副交感神經調節下降與交感神經活性增加之趨勢。本研究認為，姿勢挑戰的價值未必在於提升統計相關強度，而在於使低負荷狀態下不易觀察之自律神經反應模式更具生理可解釋性。探索性分析顯示，姿勢挑戰後出現 HRV 下降者較可能呈現與焦慮症狀相關之自律神經反應特徵，而較高焦慮症狀程度者亦傾向出現較明顯之 HRV 下降，顯示姿勢誘發之 HRV 變化可能作為反映焦慮症狀的早期生理訊號。氣泵式行動檢測儀亦展現於動態條件下穩定取得 HRV 訊號之潛力，支持未來於真實情境中進行行動式自律神經檢測與焦慮風險篩檢之可行性。; Heart rate variability (HRV) is a widely used non-invasive indicator of autonomic regulation. However, previous studies have primarily focused on resting-state measurements, with limited investigation of autonomic responses under physiological challenge conditions. Therefore, this study employed a pressure-based mobile cardiovascular monitoring device to investigate the associations between autonomic nervous system (ANS)-related parameters and anxiety and depression measures under different postural conditions, and to evaluate whether postural challenge provides additional physiologically meaningful information beyond conventional resting-state assessment. The final dataset consisted of 90 participants. Levels of anxiety and depressive symptoms were evaluated using the GAD-7, ASI-3, PHQ-4, and BDI-II questionnaires. HRV indices, including HRV, low-frequency (LF) power, high-frequency (HF) power, and the LF/HF ratio, were measured under sitting and standing conditions. The results demonstrated that overall HRV significantly decreased from 40.41 ± 4.34 in the sitting condition to 38.82 ± 6.33 in the standing condition (p = 0.014), indicating that postural transition successfully elicited measurable autonomic adjustments. Although associations between HRV and psychological measures were generally weak, autonomic response patterns observed during standing appeared to align more closely with previous findings regarding anxiety- and depression-related autonomic dysregulation. Higher psychological symptom scores tended to exhibited lower HRV and HF measures and relatively higher LF-related indices, suggesting reduced parasympathetic modulation and increased sympathetic dominance. These findings suggest that the value of postural challenge may lie not in strengthening statistical associations, but in revealing autonomic response patterns that are more physiologically interpretable. Furthermore, exploratory analyses suggested that posture-induced HRV reduction may represent an early physiological signal associated with elevated anxiety symptoms. Individuals demonstrating reduced HRV during postural challenge appeared more likely to exhibit anxiety-related autonomic characteristics, highlighting the potential value of pressure-based mobile HRV monitoring in future anxiety screening applications. 2026-01-01T00:00:00Z http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/103810 標題: 基於注視引導與知識蒸餾進行具有資源效率的胸部X光報告生成; GAZE-GUIDED KNOWLEDGE DISTILLATION FOR RESOURCE-EFFICIENT CHEST X-RAY REPORT GENERATION 作者: Nandhitha Suruliandi; Nandhitha Suruliandi 摘要: 儘管胸部X光報告生成技術已有長足進展，效能最佳的系統多半體積龐大、運算資源需求高，在放射科醫師人力最為短缺的地區反而最難取得。本研究旨在探討能否訓練一個輕量模型，使其在較大模型的監督下學習，並在不需專用圖形處理器的情況下產生具臨床價值的報告，同時釐清此種臨床價值應如何衡量。 為此，本論文提出一套以師生架構為基礎的注視引導知識蒸餾框架。教師模型採用具四十億參數的醫學視覺語言模型 MedGemma。其內部特徵事先由胸部X光影像萃取並儲存，因此教師模型本身不更新參數，於學生模型訓練過程中亦無需載入。學生模型參數量約五千萬，並非教師模型的縮減複本，而是一個獨立訓練、學習以與教師相同方式編碼影像的網路。除影像特徵之外，學生模型同時學習放射科醫師閱片時所注視的區域（由眼動追蹤記錄取得）以及參考報告內容。訓練完成後，學生模型可在不依賴教師模型的情況下獨立產生報告，其規模適用於硬體受限的部署環境。 實驗結果顯示，學生模型雖遠小於教師模型，但在本領域常用的文字重疊指標（BLEU、METEOR、ROUGE-L 與 CIDEr）以及臨床準確度上皆優於其教師模型。其中注視引導為關鍵因素，納入注視引導後臨床準確度約提升一倍。本研究並指出，僅以自然語言生成指標並不足以判斷所產生的報告是否具臨床可靠性，因此另採用逐項發現層級的指標作為互補評估。; Although the generation of chest X-ray reports has come a long way, the most sophisticated systems are large, resource intensive and least accessible in the regions where there are few radiologists. The goal of this work is to evaluate the feasibility of developing a compact model which can be trained under the supervision of a much larger model and produce clinically useful reports without the need for a dedicated GPU, and to determine how that clinical usefulness should be measured. In this respect, this thesis introduces a gaze-guided knowledge distillation framework based on the teacher-student design. MedGemma is a 4 billion parameter medical vision-language model that serves as the teacher. Its internal features are extracted from the chest X-rays once in advance and stored, so the teacher never gets updated and is not required during student training. The student is a smaller network of the order of 50 million parameters; the student isn't a reduced copy of the teacher, but a separate model trained to encode each image in the same manner as the teacher. It also trains on the areas a radiologist would examine when reading a chest X-ray, as recorded by eye tracking, and on the reference reports. Once trained, the student writes reports without the teacher, at a size suitable for hardware-limited situations. Although it is much smaller, the student is better than its own teacher on all the standard text-overlap metrics in this field (BLEU, METEOR, ROUGE-L and CIDEr), and on clinical accuracy. Experimental results show that gaze is the key, with clinical accuracy doubling when gaze guidance is included. It is also reported that NLG metrics alone are not sufficient for judging whether a generated report is clinically reliable, and hence per-finding metrics are used in this work as a complementary measure. 2026-01-01T00:00:00Z http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/103577 標題: 基於L3影像特徵與機器學習之大腸直腸癌肝轉移患者生存預測：隨機生存森林模型建構; A Random Survival Forest Model for Predicting Survival in Patients with Colorectal Liver Metastases Using Third Lumbar Vertebra Imaging Features and Machine Learning 作者: 劉珈均; Jia-Jyun Liu 摘要: 大腸直腸癌肝轉移（Colorectal Liver Metastases, CRLM）患者預後異質性高，現行 Fong Clinical Risk Score 僅整合腫瘤負荷指標，未涵蓋骨骼肌、脂肪與骨骼健康度等宿主因素；傳統第三腰椎（Third Lumbar Vertebra, L3）體組成分析多依賴人工標註，難以於臨床落實。本研究自 The Cancer Imaging Archive（TCIA）Colorectal-Liver-Metastases 納入完整案例 168 名患者（排除 CRS 缺失者）作為訓練與內部驗證集，並另取臺大醫院（National Taiwan University Hospital, NTUH）22 名患者為外部驗證集；L3 切面以 TotalSegmentator 自動定位，肌肉與脂肪之多組織分割由既有深度學習身體組成模型產出，並另整合 TotalSegmentator 之 L3 椎體遮罩以量化椎體骨骼（面積與骨骼平均衰減值）及串接以骨骼肌指數為基礎之肌少症自動辨識，生存預測以隨機生存森林（Random Survival Forest, RSF）為主並與 XGBoost Survival Embeddings（XGBSE）於五折交叉驗證下對照，以 SHAP（SHapley Additive exPlanations）提供可解釋性。自動化流程於數十秒內完成單例處理，RSF 於 TCIA 訓練／內部驗證集 C-Index 為 0.9073 ／ 0.7333、外部驗證集 0.7261、五折平均 0.7163 ± 0.013，略優於 XGBSE（0.7013 ± 0.014），內部驗證集 Kaplan-Meier 達 log-rank p = 0.0262，SHAP 顯示臨床風險評分（CRS）、年齡、皮下脂肪面積與新增之椎體骨骼面積為主要貢獻特徵，骨骼平均衰減值等亦進入前段，且內、外部驗證集之主導特徵高度一致，外部驗證集（n=22）則初步顯示與內部驗證一致之區分趨勢。本研究主要成果為建立一套可重現之自動化 L3 體組成分析流程，並以 CRLM 為示範場景初步驗證其方法學可行性；影像端已封裝為桌面端軟體，新增之椎體骨骼量化填補既有工具缺口，初步生存預測結果為「L3 體組成結合機器學習」於癌症預後評估提供方法學可行性證據。受限於樣本規模，本研究之生存預測結果尚不足以支持直接之臨床部署宣稱。; Prognostic heterogeneity in Colorectal Liver Metastases (CRLM) is poorly captured by tumor-burden tools such as the Fong Clinical Risk Score, which omit host factors such as muscle, adipose and bone health, while manual analysis at the third lumbar vertebra (L3) for body composition is difficult to deploy clinically. In this study, 168 complete-case CRLM patients (after excluding cases with missing CRS) from The Cancer Imaging Archive (TCIA) Colorectal-Liver-Metastases dataset were used for training and internal validation and 22 patients from the National Taiwan University Hospital (NTUH) served as the external cohort; L3 was localized with TotalSegmentator, multi-tissue segmentation of muscle and adipose tissue was performed by an existing deep-learning model, and vertebral-bone metrics (bone area and mean CT attenuation) were obtained by quantifying the TotalSegmentator L3 vertebra mask, with an automated sarcopenia classifier appended, while survival prediction used Random Survival Forest (RSF) benchmarked against XGBoost Survival Embeddings (XGBSE) under five-fold cross-validation, with SHapley Additive exPlanations (SHAP) for interpretability. The pipeline processed each case within tens of seconds, RSF reached C-Index = 0.9073 (training), 0.7333 (internal test) and 0.7261 (external test) with a five-fold mean of 0.7163 ± 0.013 slightly outperforming XGBSE (0.7013 ± 0.014), internal Kaplan-Meier stratification reached log-rank p = 0.0262, SHAP identified the Clinical Risk Score, age, subcutaneous adipose area and the newly added vertebral bone area as leading contributors (with mean bone attenuation also in the upper tier) and the dominant features were highly consistent between the internal and external cohorts, and the small external cohort (n=22) showed a preliminary discriminative trend consistent with internal validation. The main contribution is a reproducible automated L3 body-composition workflow demonstrated here on CRLM as the showcase scenario; the imaging components are packaged as a desktop application, the added vertebral-bone quantification fills a gap in existing tools, and the exploratory survival-prediction results provide methodological feasibility evidence for combining L3 body composition with machine learning in cancer prognostication. 2026-01-01T00:00:00Z