Study On Detection Techniques Of Pathological Images Of Cancer Cells

With the development of computer technology,technologies of image processing and pattern recognition are widely used in the field of cytopathological image detection,and have become an important research direction of molecular biotechnology as well as modern medicine.Cancer,as one of the major diseases that endanger human life and health,for which the early diagnosis is the key to treatment.The use of computer-aided cytological diagnosis technology can effectively reduce doctors' work intensity and improve the accuracy of diagnosis.Due to the complex background of cell images,as well as the morphological diversity and overlapping problem of cells,the computer detection and recognition is relatively more difficult.To meet clinical needs,starting from the actual situation,this paper engages in the study on the detection technology of cell microscopic images.Based on the previous research work,using microscopic images of cervical cells as experimental illustration,this paper conducts a comparative analysis for typical medical image segmentation techniques,and has made new technological improvements,in which the corresponding technical solutions are presented for the non-overlapping cell images and overlapping cell images,respectively.The main working contents of this paper are as follows,Firstly,for the nuclear regions of non-overlapping cells images,a segmentation algorithm is proposed to automatically obtain the initial seed growth point by means of threshold segmentation and joint regional growth,thus avoids the manual localization for the starting point.The algorithm uses the combination of threshold segmentation and region growing,in order to achieve the extraction of multiple nuclear regions in cell images.Then,for the cell body regions of non-overlapping cells images,the marker-controlled watershed algorithm can be used to quickly achieve targeted regional extraction.The algorithm improves the cell gradient image through internal and external markers primarily,then uses watershed algorithm for segmentation.Thirdly,for the cell body regions of overlapping cells images,this paper uses a snake model segmentation algorithm with higher accuracy which completes the extraction of target regions by positioning the initial contour points manually and fitting the overlapping cell body's boundaries through contour curves.Lastly,combined with the image processing method proposed above,this paper designs and realizes cell images detection system.The experimental results show that the segmentation algorithms proposed in this paper have achieved good results under the corresponding application scenarios.Compared with traditional methods,this method leads to a more accurate targeted regional extraction and possesses higher practical value.