16444

14644

كارشناسي ارشد

مهندسي پزشكي- بيوالكتريك

اصفهان : دانشگاه صنعتي اصفهان

1399

چهارده، 110ص. : مصور(رنگي)، جدول، نمودار

مريم ذكري، حسين رباني

محمد مير محمدصادقي

1400/02/25

كتابنامه

برق

مهندسي برق و كامپيوتر

1400/02/28

2687892

Detection of Cystoid B Scans in Optical Coherence Tomography Images using Hidden Markov model and AlexNet Niloofar Moosavi niloofar moosavi74@ec iut ac ir February 24 2021 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language Farsi Supervisors Prof Maryam Zekri mzekri@iut ac ir and Prof Hossein Rabbani h rabbani@med mui ac ir Abstract The retina is a thin light sensitive membrane that has a multilayered structure located at the back of the eyeball Theretinal diseases are various Age related macular degeneration AMD and diabetic macular edema DME are the mostcommon diseases in retina Imaging from layers and blood vessel structures of retina is one of the most important methodsused for diagnosis and treatment One of the most important retinal imaging techniques is optical coherence tomography The main goal of retina image processing is to create a retina image analysis system for manifesting retina diseases and itscorresponding symptoms such as Fluid cyst regions which are one of the main symptoms for AMD and DME diseases The main contribution of this thesis is detection of images that have cyst in AMD and DME images OCT systems gives alarge volume of images for each patient but in cases with AMD or DME all the images may doesn t have cyst and if thedoctors use them for diagnosis they consider them as normal retina images while these cases have AMD or DME Basedon experience the OCT system s operator non automatically selects a number of images from each patient s data volumeand delivers them to the ophthalmologist for diagnosis Some of these images are affected by noise and motion artifactsor parts of retina that do not have cyst So these images are not suitable for diagnosing AMD and DME diseases In thiswork a new method to detect cystoid B scans in OCT volumes are presented For cystoid B scans detection in DME orAMD first a number of local feature extracting algorithms is used to extract features vectors from images In the next stepSVM classifier is used to classify B scans in two classes classes s labels are cystoid and Non cystoid In the third stepthe most appropriate local feature extracting algorithm that makes the most difference between the two classes is selected This algorithm is Histogram of Oriented Gradient In the next step the feature s vectors that is extracted from Images byHOG algorithm is given to Hidden Markov model as observation vector to find parameter of this model and compared withfeatures that extracted from AlexNet by transfer learning and is given to same markov model This Hidden Markov Modelhad two hidden States cystoid and non cystoid In the last step the most probable state sequence is estimated by Viterbialgorithm Finally Each image will have a state that Determines it s class As the result SVM classifier and Hidden MarkovModel performance is compared It is observed that the Hidden Markov Model with features extracted by transfer learninghas better performance than SVM classifier For this purpose all algorithms have been implemented in MATLAB

مريم ذكري، حسين رباني

محمد مير محمدصادقي