| 摘要: | 隨著醫療技術的持續進步，再生醫學已逐漸成為現代醫學研究的重要發展方向。幹細胞憑藉其自我更新與多向分化能力，在組織修復、免疫調節與細胞替代等應用中展現極大潛力。其中自體幹細胞因具備高度免疫相容性與無排斥反應等優勢，已成為臨床再生治療的關鍵來源。然而自體幹細胞在體外培養過程中常因培養環境與體內微環境差異過大，如缺乏機械刺激、過度暴露於高氧環境、或長時間傳代等因素，導致細胞功能退化、活性下降。若無法即時掌握細胞在培養期間的功能狀態，可能導致低活性的細胞被誤用於臨床，進而影響治療成效，甚至導致資源浪費。
為解決自體幹細胞培養過程中活性監測的問題，本研究聚焦於自體幹細胞的運動活性（motility）評估，開發一套結合深度學習與電腦視覺技術的細胞自動追蹤與分析系統。系統採用 YOLO11 模型進行自體幹細胞目標偵測，並結合 BoT-SORT 多目標追蹤演算法，能於顯微影像中即時擷取每一幹細胞的移動軌跡及時間資訊。本研究設計「細胞運動活性指標」（Cell Motility Index, CMI），以細胞在特定時間區間內的速度、位移距離與方向穩定性作為核心參數，量化幹細胞的運動能力，間接反映其運動活力與遷移潛能。為了進一步評估整體自體幹細胞群體的運動活力，本研究對 CMI 數據進行統計建模，並計算其分佈參數 μ 與 σ。基於此模型，定義細胞運動活力指標 motility = μ/σ，該指標綜合反映了自體幹細胞群體運動活性的平均水平與變異程度，提供更全面且穩健的活性評估依據。
與傳統以化學螢光標記或奈米材料進行細胞活性觀測的方式相比，本系統採取完全非侵入式架構，避免標記造成的毒性與訊號衰減問題，並具備即時性、高解析度、低成本與可長時間觀察等優勢。最重要的是，本系統能協助研究者快速判斷不同試管或培養條件下的自體幹細胞運動活性變化，作為自體幹細胞品質控制與回輸決策的重要依據，避免「幹細胞看起來健康但實際活力不足」的情況，提升整體自體幹細胞治療的效益與可靠性。
本研究所提出的運動活性分析方法，開啟了以「動態行為」評估幹細胞狀態的新方向，未來亦可應用於不同類型幹細胞的行為研究，具備高度擴展性與臨床轉譯潛力。
;With the continuous advancement of medical technology, regenerative medicine has gradually become a vital focus in modern medical research. Stem cells, owing to their abilities for self-renewal and multilineage differentiation, demonstrate great potential in applications such as tissue repair, immune regulation, and cell replacement therapies. Among them, autologous stem cells are especially important in clinical regenerative treatments due to their high immune compatibility and absence of rejection responses. However, during in vitro cultivation, autologous stem cells often suffer from decreased functionality and diminished activity caused by significant differences between the culture environment and the physiological microenvironment. Factors such as lack of mechanical stimulation, prolonged exposure to high oxygen levels, or extended passaging contribute to this decline. Without timely monitoring of cellular functional status during cultivation, low-activity cells may inadvertently be used in clinical applications, which could compromise therapeutic outcomes and lead to resource wastage.
To address the challenge of monitoring autologous stem cell activity during cultivation, this study focuses on evaluating the motility of autologous stem cells by developing an automated cell tracking and analysis system that integrates deep learning and computer vision technologies. The system utilizes the YOLO11 model for autologous stem cell detection and combines it with the BoT-SORT multi-object tracking algorithm, enabling real-time extraction of individual cell trajectories and temporal information from microscopic images. We further designed a Cell Motility Index (CMI) that quantifies cellular motility based on core parameters including speed, displacement distance, and directional persistence within a defined time window, indirectly reflecting the physiological vitality and migratory potential of the cells.
To comprehensively evaluate the overall motility of autologous stem cell populations, we performed statistical modeling on the CMI data and calculated distribution parameters μ and σ. Based on this model, we defined a motility metric as motility = μ/σ, which integrates both the average motility level and variability across the cell population, providing a more robust and comprehensive assessment of cellular activity.
Compared to traditional methods relying on chemical fluorescence labeling or nanomaterials for cell activity observation, our system adopts a completely non-invasive framework that avoids toxicity and signal decay associated with labeling. It offers advantages of real-time monitoring, high resolution, low cost, and long-term observation capability. Most importantly, the system enables researchers to rapidly assess changes in cell motility across different culture tubes or conditions, serving as a critical basis for quality control and reinfusion decisions. This helps prevent scenarios where cells appear healthy morphologically but have insufficient functional activity, ultimately improving the efficacy and reliability of autologous stem cell therapies.
The motility analysis approach proposed in this study opens a new direction for evaluating stem cell status based on dynamic behavior. It is also expected to be applicable to the behavioral studies of various stem cell types, possessing high scalability and translational potential in clinical applications. |
