Research And Implementation Of Mammographic Image Tumor Segmentation By Extreme Learning Machine

Breast cancer is one of the common female malignant tumors.In China,it has become the second killer of cancer for women.But it's reported that early detection of breast tumor can save lives effectively.And mammography has become one of the main ways to detect the tumor because of its low radiation and low cost.Unfortunately,due to the human factors,such as fatigue,time consuming,misdiagnosing and so on,universal detection of breast cancer is difficult to implement.Then computer aided diagnosed system for tumor detect is required,which can improve accuracy and efficiency.What's more,it's acknowledged that image segmentation is a difficult process during total medical image recognition and image diagnosis.Therefore,this paper proposes a breast tumor segmentation algorithm based on Extreme learning machine,which is called ELM for short.Mammographic image contains noise and has low image gray contrast.As a result,image preprocessing is needed.Here median filter and wavelet denoising is used to eliminate the image noise.During the enhancement,two ways have been implemented and contrastive analyzed.One way is based on the space domain.Adaptive histogram equalization,morphology enhancement is used during this process.The other way is to process in the contourlet domain.Image should be transformed into contours wavelet and enhance the gray level contrast between tumor and other structure.After denoising and enhancement,region of interest will be attracted based on the gold standard achieved from the radiologist.Then the samples are separated into training set and testing set.Finally,the segment result will get by ELM.During the evaluation validation process,feature vectors of segmentation result,such as gray co-occurrence matrix,accuracy,will be extracted,which are used to contrast with those of golden standard.SVM algorithm is used as a comparison to evaluate the performance of the ELM.In conclusion,used as a classifier to segment on the mammographic tumor image,ELM performs better than SVM.