2024-01-012024-05-13https://scholars.lib.ntu.edu.tw/handle/123456789/652725"隨著老年人口的增加，失智症人口也快速增加，使得失智症成為一個嚴重的健康與社會問題。失智症是一種神經退化疾病，疾病使腦部結構及功能病變，進而造成失智患者認知、功能及行為出現漸進式退化。雖然腦部的退化無法治癒，早期偵測認知退化將有助於提供適當醫療處置或復健介入，進而增加失智症患者之生活品質。早期偵測系統之建立，也有助於發展認知退化患者之精準醫療模式。腦部的退化會使過去習得的動作技能受到影響，因此書寫可能是偵測認知退化一個有希望的行為指標，相較於目前用來早期偵測認知退化的影像及生化指標而言，書寫任務更便宜且更容易在生活日常執行。此外，人工智慧可以透過學習臨床專家判斷疾病的模式，適合用於建立早期疾病偵測系統。因此，本研究的目的旨在利用書寫任務作為行為標記，發展一認知退化老人之人工智慧早期偵測系統。 本計畫為一個兩階段的研究，在第一階段中，我們將建立一套針對認知退化病人使用之平板化中文書寫評估應用程式。為了提升此應用程式之臨床可用性，我們將同時驗證此應用程式之心理計量特性，我們招募了20位健康老人及20位阿茲海默症患者，阿茲海默症是失智患者中最大量的族群，因此本研究選阿茲海默症患者作為失智症的代表，我們將利用40位個案表現建立此應用程式之再測信度及已知族群效度。本計畫第二階段將利用書寫任務作為行為標記發展一認知退化老人之人工智慧早期偵測系統。我們將招募40位健康老人、40位主觀認知退化、40位輕度認知障礙及40位阿茲海默症患者，每一位患者將利用第一階段發展之應用程式評估書寫表現，每一位患者將評估四次，包含初評、六個月、一年及一年半之追蹤評估。我們將分析包含準確性、反應時間、易讀性、書寫運動學及筆尖與握筆壓力之書寫表現變數，比較健康老人、主觀認知退化、輕度認知障礙及阿茲海默症患者之書寫表現差異，可區辨正常老人及病人或可區辨不同認知退化族群之書寫變數則為認知退化疾病患者早期偵測因子。這些早期偵測因子將作為建立人工智慧系統之候選特徵，不同受試者族群之資料及同一受試者之長期追蹤數據將輸入人工智慧模型進行訓練。我們將使用決策樹模型，因為此類模型具有較好的自我解釋力，同時可模擬醫生做決定時的想法，而且，此模型背後的機制是比較容易被了解的。最後我們將選出最好的人工智慧模型，建立認知退化老人之人工智慧早期偵測系統。 本計畫將利用書寫任務作為行為標記，發展一認知退化老人之人工智慧早期偵測系統，此系統便宜且易於在診間或日常生活中執行，除了早期偵測，此系統亦可作為疾病進程評估及醫療或復健計畫療效評估指標，此將有助於發展認知退化患者之精準醫療模式，降低醫療及社會經濟成本。" "As the increase of the world’s elderly population, patients with dementia is rapidly rising, making dementia an increasing serious health and social problem. Dementia is one kind of neurodegenerative diseases, which affect the structure and function of brain regions, resulting in a progressive cognitive, functional and behavioral decline. Although there is no cure for the signs of brain degeneration, early detection of cognitive decline would be crucial for providing appropriate medical treatment and rehabilitative intervention, which help improving the quality of life for patients with cognitive decline. It is also helpful for developing precision medicine of patients with cognitive decline. Handwriting is a promising behavioral marker for early detection of cognitive decline since changes in brain result in impairments of performance of previously learned motor skills. Compared to the laboratory and brain-image tests, evaluation of handwriting is low-cost and easier to implement in daily living. Besides, AI technology can learn to see patterns similarity to the way clinicians see. Therefore, the purpose of the present study is to use handwriting as a behavioral marker to develop an AI-based early detection system for patients with cognitive decline. The proposed project will be a two-phase study. In the first phase, a tablet-based Chinese handwriting evaluation APP for patients with cognitive decline will be design In order to improve the clinical utility, psychometrics of this APP will be developed. Twenty healthy elderly and 20 patients with Alzheimer’s disease (AD), the most common form of dementia, will be recruited. Test-retest reliability and known group validity will be performed. The second phase of this study is to use handwriting as a behavioral marker to develop an AI-based early detection system for patients with cognitive decline. Forty healthy elderly, 40 patients with subjective cognitive decline (SCD), 40 patients with mild cognitive impairment (MCI) and 40 patients with AD will be recruited. For each participant, handwriting evaluation using our handwriting APP developed in the first phase will be performed 4 times (i.e. first time, 6-,12-,18-month follow-up). Variables of handwriting performance, namely accuracy, reaction time, legibility, handwriting kinematics as well as tip and grip pressure will be analyzed. Differences of these variables among healthy elderly, SCD, MCI and AD will be investigated. Variables which can differentiate normal from patient groups or patient group with different severity of cognitive decline will be chosen as the early detectors of patients with cognitive decline. The early detectors will be served as the candidate feature. Cross-sectional data among different groups and longitudinal data for the same participants will be imported and trained for the AI models. Tree-based machine learning models will be used because they are self-explanatory and mimic how physician thinks while making a decision and the logic behind the tree can be understood. Finally, the best model will be selected to use in the AI-based early detection system for patients with cognitive decline."書寫認知退化老人早期偵測人工智慧handwritingcognitive declineelderlyearly detectionAI