Classification Of Malignant And Benign Breast Lesions Based On Ultrasound Video Sequence Analysis

Breast cancer is the most common cancer among women; its harmfulness hasnot be overlooked. According to medical clinical experience, if most of the breastcancer could be detected early, the treatment success rate will be greatly enhanced.The hospital ofen uses the ultrasound imaging for the diagnosis of breast cancer inChina. Computer-aided diagnosis system is used to help doctors diagnose benignand malignant tumors. Most of the existing computer-aided diagnosis systemanalysis patient’s cancer based on one single frame of the patient’s video sequence,though it will reduce the time of computer diagnosis, it still causes two mainproblem, first, how to select the single frame from the video sequence; second,different doctor will select a different image based on their different diagnosticcriteria which will affect the result of computer diagnosis. Therefore, we study howto classify the benign and malignant breast tumors based on video sequence.Most of the existing computer-aided systems get features from the tumormorphological or characteristics of the texture. This paper first describes how toextract the tumor morphology and texture feature in video sequence, then this paperpresents a normalized distribution variance sequence feature extraction methodbased on chain code.In sequence feature extraction method based on morphological, it firstlyextract10morphological characteristics that is roundness,aspect ratio,the average ofthe normalized radial length,the standard deviation of the normalized radiallength,the entropy of the normalized radial length,area ratio,roughness,lobularindex,needle-like degree,azimuth,from each frame in the sequence, then this paperproposed three sequence feature extraction operator, so we can get three sequencefeature for every morphological feature. These three sequence feature operator aresequence average characteristic,sequence second-order differential characteristic,sequence distribution entropy,they do not only take the similarity in morphologybetween adjacent frames into consideration, but also consider their differences.This paper develop the GLCM to GLCMEX so that it can be used to extracttexture features in video sequence, the commonly GLCM often used in thesingle-frame to extract texture features. The difference between the commonlyGLCM and GLCMEX is that the former only consider the horizontal and verticaldisplacement between pixels, but GLCMEX also consider the frame numberdifference.This paper also extract sequence characteristics from the tumor contour coding,the previous feature extraction method study the whole morphologicalcharacteristics of the tumor, but this paper is used Freeman Chain Code to encoding the contour of the tumor. As the experimental result show that this method’srecognition rate is96.1%, it can provide some suggestion for doctors in clinicaldiagnosis.