Research On Automatic Capture And Feature Recognition For TCM Tongue Image In Chronic Kidney Disease

In recent years,the prevalence rate of chronic kidney disease(CKD)is increasing,and the diagnosis and treatment of chronic kidney disease becomes a major medical problem.Diagnosing the kidney dysfunction early can control the development of CKD effectively.Traditional Chinese Medicine(TCM)doctors believe that the change of tongue reflects the pathological changes of patients' viscera,so that doctors can find the kidney dysfunction timely through the patients' tongue which plays good function in the diagnosis of CKD.In this thesis,tongue image automatic acquisition,feature recognition and disease prediction are studied based on the tongue image samples of patients with CKD,and a system prototype of tongue diagnosis for CKD is realized.The specific work is as follows:1.The tongue acquisition device is designed based on the clinical tongue observation method and Chinese adult statistics data to provide a standard lighting environment.And a tongue automatic detection method is proposed based on HSV space to realize the tongue image real-time detection,and the tongue image fuzzy detection is realized by extracting the image features and random forest algorithm to ensure the availability of tongue samples.2.Tongue color,tooth-marked,tongue crack,tongue spot,tongue coating color and thickness are studied from tongue image of clinical CKD.The color recognition of tongue and tongue coating is realized by random forest algorithm based on the color histogram,color moment features and the 12 color classification centers which are selected based on the tongue image color category and the distribution in CIExy color space.And their recognition accuracy is 80.8%,82.7% and 86.5%.A tooth-marked recognition algorithm is proposed by extracting three boundaries,tooth-marked areas and tooth-marked features,then the severity degree of tooth-marked is identified by random forest algorithm,and the recognition accuracy is 71.1%.A tongue crack recognition algorithm is proposed by tongue crack areas and crack features.The tongue crack areas are extracting by SLIC super pixel segmentation and region growth.The crack features are calculated according to tongue crack areas.Then the severity degree of tongue crack is identified by random forest algorithm,and the recognition accuracy is 73.1%.A tongue spot recognition algorithm is proposed based on the tongue spot areas and tongue spot features.The tongue spot areas are extracted by LOG algorithms,and tongue spot features are calculated from spot areas.Then the severity degree of tongue spot is identified by random forest algorithm,and the recognition accuracy is 76.9%.Then the thickness of tongue coating is identified by random forest algorithm and the tongue coating features from the nine image blocks by gray symbiotic matrix and tongue coating rate,and the recognition accuracy is 73.1%.Finally,chronic kidney disease prediction is studied based on the theory of TCM syndrome differentiation and the tongue features identified above to assist the diagnosis.3.The decision support system prototype for tongue diagnosis of CKD is designed and implemented by C# and WPF.The system includes four modules: tongue image acquisition,tongue image segmentation,tongue image features recognition and disease prediction.The acquisition module is used to capture and import tongue image.The segmentation module is used to extract tongue body.The features recognition module is used to recognize tongue features,and the disease prediction module is used to predict syndromes based on tongue features.