Dictionary Transfer Learning Based Stomach CT Image Sequences Co-segmentation

With the rapid development of technology, medical imaging equipments becomemore and more advanced, and it makes the acquisition of medical imaging more andmore easily. As people have been much more concerned about their own health, thenumber of hospital radiology images have increased exponentially, which greatlyaggravates the burden on doctors and makes it more difficult for doctors to guaranteetheir diagnostic accuracy. To solve this problem, this paper extracted lymph node regionof interest from stomach CT images, mainly to complete the following work:1. Dictionary learning based co-segmentation method is proposed. Thecharacteristics of gastric lesions area in CT images are fuzzy boundaries and smallgrayscale differences in peripheral tissues and organs, so accurate segmentation resultsare got by using interactive segmentation method for a small amount of image, and thenbased on the results to learning dictionary and match the rest of the image sequence toget result, which ensure the accuracy and save time.2. Dictionary transfer based co-segmentation is proposed, which combination withdictionary learning and transfer learning. The dictionary is updated by transferring thelarge changes in the object regions in the CT image sequence to enhancing itsdescription and completeness, thus the segmentation result of inaccurate is avoided forthe change between the image sequences.3. A fast interactive dictionary selected based co-segmentation method is proposed.Real-time and accuracy are needed by computer-aided diagnosis, so the pixel path isextracted instead the features, and the accuracy and robustness of our algorithm isimproved by using interactive segmentation again for the incorrect segmentation result,which avoiding the incorrect segmentation for individual differences in CT images ofdifferent person.