| 摘要: | 舌診在中醫（TCM）中起著不可估量的作用，是中醫主要的診斷方法之一。通過中醫師以目視檢查舌頭及其特徵（包括活力，形狀，顏色，舌表層和水分）來進行檢查。這些信息本身或與舌頭影像圖譜的映射都可以幫助中醫醫生識別並確定患者體內的疾病。但是，舌診是根據醫生的知識和經驗來完成的，容易缺乏客觀的評估標準而可能會阻礙舌診的應用和有效性。隨著計算能力的提高和圖像處理技術的發展，技術上已經能在醫院中開發並使用了自動量化舌頭各層面的診斷設備。在這些技術優勢的基礎上，我們開發了一種基於人工智慧演算法的舌頭圖像辨識分割方法，該方法可以較容易辨識與分析目標舌頭的形態特徵，而不受時間和空間的限制。然而，為了實現這些任務，當代的人工智慧演算法遭受了諸如舌頭位置的干擾，光損傷或光污染以及色彩校正等困境。為了克服這些情況，我們通過將舌頭區域的長寬比和嘴唇的長度引入該方法，重新設計了傳統的舌頭圖像獲取方法。本方法可通過自動計算縱橫比來幫助用戶獲取正確的舌頭照片。然後，本論文將進一步採用Retinex演算法校正光的影響，包括光源的方向和每個光矢量的強度。因此，通過合併來自光向量的方向和強度的光信息，可以修改照片上RGB像素的強度。最終，將採用卷積數據密度泛函理論進行特徵提取和圖像分割。在臨床研究中，希望將來能從Landseed國際醫院的中醫診所收集大量舌頭照片，以建立雲數據庫，相關參與者和專家將幫助為監督的機器學習算法標記圖像。這項研究的期望結果是提供一種簡單的方法來連接舌頭形態和中醫應用。;Tongue diagnosis plays an invaluable role in Traditional Chinese Medicine (TCM) and is one of the major methods of diagnosis. It is performed by visual inspection of the tongue and its features, including vitality, shape, color, coating, and moisture. These pieces of information alone or mapping with the topographic regions of the tongue help a TCM doctor to identify and locate the disease in a patient’s body. However, tongue diagnosis is achieved based on the knowledge and experiences of a doctor, and the lack of objective evaluation standards may hinder the application and validity of tongue diagnosis. With the improvement of computing capability and the development of image processing technology, diagnostic equipment that automatically quantifies various aspects of tongues has been developed and used in hospitals. Under the foundation from these technical benefits, we developed a method of tongue image analysis based on artificial intelligence algorithms that allow easy monitoring and recognizing the morphological features of the targeted tongue without time and space limitations. However, in order to achieve these tasks, the contemporary AI algorithms suffer the predicaments, such as the disturbances of tongue position, light damage and/or pollution, and color correction, and so on. To overcome these circumstances, we redesign the conventional method of the tongue image acquisition by introducing the aspect ratio of the tongue area and length of lip into the method. The APP on the mobile phone will help users acquire correct tongue photos by calculating the aspect ratio. Then the Retinex algorithm will be employed to correct the effect of light, including the direction of the light source and the intensity of each light vector. Thus, the intensity of the RGB pixels on the photo would be modified by merging the light information from the directions and the intensities of the light vectors. On the clinical investigations, a large set of tongue photos will be collected from the TCM clinic of Landseed International Hospital to establish a cloud database, and relevant participants and experts will help label the images for the supervised machine learning algorithm. Eventually, the convolutional data density functional theory will be adopted for the feature extraction and image segmentation. The desired outcome of this study is to provide an easy methodology to connect the tongue morphologies and the TCM applications. |
