Study On Pathological Cell Aided Detection Based On Machine Learning

CAD is the use of computers to learn from the experience of human beings,to achieve the automatic diagnosis of the disease.With the development of machine learning technology,the field of application has been expanded,and the computer aided diagnosis combined with medical image has become a research hotspot.It is the goal of scientific and technical personnel to reduce the workload of doctors and improve the accuracy of diagnosis by automated auxiliary detection.By using a large number of cell images produced by the hospital every day,combined with the idea of image correlation technology and machine learning,the classification model of cell aided detection was established.Because of the difference of the cell image and the common medical image,it has the irregularity and the more interference factors,which leads to the difficulty of recognition.Due to the large differences in various types of cells,can not be unified research.In this paper,the cervical cell image as the main object of study,on the basis of the study of the auxiliary detection technology,including image preprocessing,image feature extraction and analysis and image classification technology.The detailed research contents include the following aspects:First of all,the segmentation based on active contour model,and C-V model level set algorithm and curve evolution theory based on the successful extraction of region of interest by C-V cell image segmentation model,the nucleus and the cytoplasm.Secondly,according to the feature extraction technology of cell image,the global features of cell image are studied.In this paper,according to the particularity of the cell image,an improved texture extraction method based on gray level co-occurrence matrix is proposed.The heuristic search algorithm is used to extract the most effective feature subset for subsequent classification.Finally,according to the classification of cell images is studied in this paper,the mainstream of machine learning techniques,including neural network and SVM,respectively to them as classifier,feature extraction from the front to as input for training,classification of experimental cervical cells.On this basis,an improved particle swarm optimization algorithm is proposed,which can optimize the weights of neural network and improve the accuracy of the classifier.