Automatic Extraction For Lesion In Special Medical Image

With the Computed Tomography(CT) continuously developing and upgrading, CT has become an important tool of the diagnosis for brain disease. Through the brain CT image to read and analyze, radiologist could make a qualitative diagnosis according to clinical experience and patient's medical history. The extensive application of CT technology brings increasing number of CT image data, so, diagnosis's workload is getting bigger and bigger. And human's subjective influence causes the difference of diagnosis, lacking objective data and increasing the risk of misdiagnosis and mistreatment. Under this situation, computer aided diagnosis is expected to solve the partial problems. We could detect the diseased area and morphology parameter by automatic extraction of lesions.This article analyzes the algorithms used widely in medical image segmentation, introducing the characteristic and the applicable scope of each kind of algorithm. Because the most lesions segmentation method is executed after getting the information of lesion type and the lesion features, it is not actual automatic segmentation for unknown lesions. Based on prior knowledge about a gross bilateral symmetry of brain, we want to implement segmentation for unknown lesions by analyzing left and right brain hemispheric texture feature differences.First, because Mid-Sagittal Line(MSL) from patient's head may present the different angle during the examination process, MSL extraction is needed in brain CT image, and then, normalized image is obtained at last. We take the mutual information as similarity measure and detect counterpart point at left and right brain hemisphere to determinate the MSL. This method not only increases the precision of determining MSL, but also develops the tolerance to asymmetry brain and robustness.Secondly, we take wavelet transformation as tool and automatic lesion extraction as goal, and then, we propose algorithm based textural feature for lesion extraction through comparison between left and right brain hemisphere textural features. After determining information of lesion location and size, we extract lesion outline for the second time by C-V level set to cause the lesion boundary be more precise. This algorithm realizes the automatic extraction for unknown lesion in brain CT image, and can obtain the precise lesion boundary.Finally, in order to overcome the shortcoming of radiologist's subjective in the process of diagnoses, the differences of diagnosis results between different radiologists to the identical case. Based on extracted lesion, we quantize the parameters of lesion's area and grayscale, and provide the objective basis for brain lesion diagnosis.