بهبود سرعت "انطباق مبتني بر روش برش گراف" جهت انطباق غير صلب تصاوير تشديد مغناطيسي مغز

Speed improvement in graph-cuts-based registration for non-rigid image registration of brain magnetic resonance images

فرآيند نگاشت نقاط از يك تصوير به نقاط متناظر آن در تصوير ديگر انطباق تصوير ناميده مي‌شود كه تبديلي مكاني است. روش‌هاي انطباق تصوير را مي‌توان به لحاظ ماهيت تبديل انجام گرفته بر روي تصوير به انطباق صلب و ناصلب طبقه‌بندي كرد. يكي از راهكارهاي انطباق تصوير ناصلب، حل مسئله انطباق بعنوان يك مسئله بهينه‌سازي است. نمونه‌اي از اين روش‌ها انطباق بوسيله برش گراف مي‌باشد. اساس اين روش بر مبناي ايجاد گرافي اختصاصي شده براي تابع انرژي مورد نظر (كه بايد بهينه شود) است، به نحوي كه برش كمينه روي اين گراف منجر به كمينه‌سازي انرژي نيز بشود. با توجه به اهميت مسئله زمان در كاربردهاي كلينيكي و پزشكي، در اين تحقيق با ايجاد تغييراتي در روشي كه از برش گراف براي انطباق تصوير استفاده مي‌كند روشي پيشنهاد داديم كه در آن زمان انجام فرآيند انطباق به شدت كاهش مي‌يابد. نتايج حاصل از پياده‌سازي روش پيشنهادي ما بر روي تصاوير تشديد مغناطيسي مغز نشان مي‌دهد كه سرعت اجراي الگوريتم پيشنهادي نسبت به الگوريتم اصلي در قبال اندكي افزايش خطا (افزايش مقدار متوسط معيار SAD از 0/7 به 1) تقريباً سه برابر مي‌شود.

Image processing methods, which can visualize objects inside the human body, are of special interests. In clinical diagnosis using medical images, integration of useful data from separate images is often desired. The images have to be geometrically aligned for better observation. The procedure of mapping points from the reference image to corresponding points in the floating image is called Image Registration. It is a spatial transform. These images might be different because they were taken at different times or applied by using different devices. By the nature of this image transformation, image registration can be classified into rigid registration and non-rigid registration. The freedom’s degree in a rigid transformation is relatively low and the methods of rigid image registration are becoming mature. In contrast, non-rigid image registration is still a challenging problem because of its high degree of freedom. One of the non-rigid image registration methods is turning the registration problem into an optimization problem and obtaining the optimal value as the result of registration. An example of these methods is the graph-cuts based registration. The basic technique is to construct a specialized graph for the energy function to be minimized in a way that the minimum cut on this graph also minimizes the energy. Given that our focus in this research, is on the medical image registration, and time is one of the critical factors in medical applications. It seems that improvement of this method in terms of run time will be helpful for its clinical and medical applications. In order to achieve this goal, in this research, with modifying the energy function, we proposed a method that significantly reduces the run time of registration process. The implementation results of our proposed method on the images with artificial deformations which are similar to the most pessimistic possible deformation modes in real image data, show that the proposed algorithm is about three times faster than the existing algorithm, while the average amount of SAD criterion will be increased from 0.7 to 1.