- 學位論文
基於手繪螺旋圖之帕金森氏症輔助診斷研究
Assisted Diagnosis of Parkinson's Disease Based on Hand-drawn Spiral Graph
摘要
帕金森氏症是第二大最常見的神經退化性疾病，病徵為多巴胺神經元死亡，導致震顫、肌肉僵硬、運動遲緩，其引起的殘疾和死亡增長速度比任何其他神經系統疾病都快。目前，帕金森氏症的診斷通常依賴於臨床表現和神經學檢查，然而，這些檢查可能存在主觀性和可變性，易導致診斷的不確定性和誤判率的提高。此外，醫師臨床診斷需要多次檢測與長期觀察，病人前往問診所需時間、次數與相關成本並不低，且相當仰賴醫師的經驗。因此，一個快速且客觀的輔助檢測工具就顯得相當需要。再者，帕金森氏症初期病徵並不顯著，病人容易漠視且誤判，而早期治療將有助於病情的延緩，若能有簡易自我檢測的方式將有助於病患自我保護，以及疾病的早期發現與治療。 許多研究顯示，手繪螺旋圖的檢測快速、簡易與客觀，有助於帕金森氏症的診斷，因此，本研究利用影像辨識和機器學習發展出一個帕金森氏症手繪螺旋圖分類模型。為了提升模型的一般化程度(Generalization)，本研究採用3個手繪螺旋圖資料集進行評估，提高模型的泛化能力。研究結果顯示，在卷積神經網路、K近鄰法、邏輯迴歸、支援向量機、決策樹與隨機森林等方法中，卷積神經網路的帕金森氏症手繪螺旋圖檢測表現最為突出，準確率達92.52%，且誤判率最低(False Negative rate)。而透過此模型的輔助診斷，將協助專業醫師提升帕金森氏症的診斷效率、減少人為判斷誤差，再者，經由手繪螺旋圖檢測方式，將協助使用者容易自我檢測，提高儘早發現與治療病症機率。
並列摘要
Parkinson's disease is the second most common neurodegenerative disorder, characterized by the death of dopamine neurons, leading to tremor, rigidity, slowness of movement. The rate of disability and mortality caused by Parkinson's disease is faster than any other neurodegenerative disorder. Currently, the diagnosis of Parkinson's disease relies on clinical manifestations and neurological examinations, which may be subjective and variable, resulting in increased uncertainty and misdiagnosis rates. Moreover, clinical diagnosis by physicians requires multiple tests and long-term observation, which can be time-consuming, frequent, and costly, heavily relying on the doctor's experience. Therefore, a fast and objective auxiliary diagnostic tool is highly needed. Furthermore, the early symptoms of Parkinson's disease may not be prominent, making it easy for patients to ignore or misjudge them. Early treatment can delay the progression of the disease, so a simple self-assessment method would help patients protect themselves and facilitate early detection and treatment of the disease. Numerous researches have shown that hand-drawn spiral graph tests are quick, easy, and objective, aiding in the diagnosis of Parkinson's disease. Hence, this research developed a Parkinson's disease hand-drawn spiral graph classification model using image recognition and machine learning. To improve the model's generalization, three hand-drawn spiral graph datasets were used for evaluation, enhancing its ability to generalize. The research results indicated that among methods such as convolutional neural network, K-nearest neighbors, logistic regression, support vector machine, decision tree, and random forest, the convolutional neural network showed the most outstanding performance in detecting Parkinson's disease hand-drawn spiral graph, with an accuracy of 92.52% and the lowest false negative rate. By using this model for auxiliary diagnosis, professional physicians can improve the efficiency of diagnosing Parkinson's disease and reduce human judgment errors. Additionally, through the hand-drawn spiral graph test, users can easily self-assess and increase the likelihood of early detection and treatment of the disease.