بازشناسي وفقي وقوع بيماري¬هاي كبدي در تصاوير سي تي: مدلي خودكار با تلفيق توصيفگرهاي ماتريس هم رخداد و مدل سلسله مراتبي پردازش بينايي

Adaptive recognition of liver disease on CT images: Automatic model by combining Co-occurrence matrix and hierarchical model of visual processing descriptors

به سبب برخي تضادها در نتايج استنباطي از تصاوير سي¬تي و تباين مشابه ميان بخش¬هاي مختلف كبد از سوي پزشكان، تشخيص دقيق بيماري¬هاي احتمالي، تبحر، تجربه، دقت و زمان مي¬طلبد. در اين مقاله، مدلي تركيبي مبتني بر تكنيك¬هاي پردازش تصوير و يادگيري ماشيني ارائه شده كه مي¬تواند در تفكيك ضايعات ناشي از بيماري كبدي در تصاوير سي¬تي موثر باشد. مراحل پياده¬سازي شامل چهار فاز است. (1) پيش پردازش تصاوير نمونه سي¬تي، بهره¬گيري از روش تبديل موجك گسسته در حذف نويز و جداسازي ناحيه مورد علاقه تصوير؛ (2) ساخت الگوي بازشناسي كه بر مبناي استخراج ويژگي¬ها توسط ماتريس هم رخداد سطوح خاكستري و توصيفگر قدرتمند Hmax رخ مي¬دهد؛ كاستن ابعاد ويژگي¬ها نيز توسط الگوريتم آناليز اجزاي اصلي و طبقه¬بندي توسط ماشين بردار پشتيبان و در نهايت (3) ارزيابي به روش K-fold در عملكرد. نتايج در پياده¬سازي چه در مرحله ارزيابي عملكرد و چه در مرحله بهره¬گيري از انتخاب ويژگي رضايت بخش است؛ دقت بازشناسي به طور متوسط در حد مطلوب 92% قرار دارد و عدم قطعيت مورد بررسي قرار گرفته است. پياده¬سازي اين سيستم در مراكز بيمارستاني و كلينيك منجر به تشخيص صحيح و بلادرنگ بيماري كبدي در حد گسترده¬اي خواهد شد.

The liver is the largest gland in the body located on the side of the stomach and the gall bladder, the organ is critical that no artificial body failed to replace it. According to the World Health Organization, liver cancer is one of the leading causes of death in the world. Liver cancer as a mass in the right upper quadrant and constitutional symptoms such as jaundice and weakness shows, in recent years has been increasingly mortality rate. Early detection and accurate staging of liver cancer is an important issue in practical radiology. Abnormal liver tissue liver lesions say. Due to some inconsistencies in analytical results of the CT images and the same contrast between different parts of the liver from doctors, accurate diagnosis possible conditions, skill, experience, accuracy and My¬Tlbd time. So even skilled doctors also provide different results this location. In this thesis, efficient solution combines the techniques of image processing and machine learning model proposed in the resolution of lesions caused by liver disease that can be effective in CT images. The implementation process includes four phases. (1) pre-processing of images CT, utilizes discrete wavelet transform method for removing image noise and isolate the area of interest; (2) the construction of pattern recognition based on gray level co-occurrence matrix and descriptor extracted by the powerful Vyzhgy¬Ha Hmax occur, reduce the size Vyzhgy¬Ha by principal component analysis and classification by SVM algorithm is optimized and finally (3) K- equal in performance evaluations. Implementation in the presence of both descriptor regardless reduce the Vyzhgy¬Ha, despite the unique descriptor Hmax and reduced dimensions and characteristics as well as descriptor applied to both carried and reduce the size and impact of each on the diagnosis measured Vyzhgy¬Ha Is. The implementation of this system in hospitals and clinics to correct diagnosis and real-time liver disease will be too wide.