| In the field of medical diagnosis and research,medical imaging provides a key auxiliary role for clinical diagnosis.However,in traditional clinical medical work,it is mainly dependent on radiologists or experts combined with their own experience to make subjective judgments on images,which is not only inefficient,but also may lead to missed and misdiagnosed situations,which can no longer meet the needs of the development of current medical field.With the continuous development of computer-aided therapy,machine learning has attracted great attention in the field of medical research in recent years,especially machine learning methods represented by deep neural networks.Therefore,in order to break through the limitations of traditional manual reading,further improve the work efficiency of radiologists and the accuracy of diagnosis,in this paper,was based on a variety of methods of machine learning the medical imaging of the meniscus and female cervical images were depth studied.The main research works and innovative achievements of this paper include the following two aspects:(1)Aiming at the localization and segmentation of meniscus in MRI of knee joint,a method based on Mask region convolutional neural network(Mask R-CNN)is designed and proposed,which can directly extract features from images to realize automatic localization and segmentation of meniscus.In order to highlight the proportion of the meniscus,the initial image data is pre-processed to reduce it to about1/8 of the original image,and then gamma transform is used to enhance the contrast of the images;then,the pre-processed image data is input into the pre-trained Mask RCNN.In this process,the transfer learning method is used to generate the weight for network training.By randomly selecting 1000 images for testing,the mean values of Intersection-over-Union(Io U)and Dice Similarity Coefficient(DSC)reached 83.68%and 91.13%,respectively.Compared with the traditional method of manual feature extraction,this method can not only ensure the quality of meniscus segmentation,but also effectively reduce manual intervention and improve the efficiency of meniscus segmentation.(2)Aiming at the research on the classification and prediction of cervical lesions in female cervix images,a new method is proposed to use the improved VGG-16 network to achieve the classification and prediction of female cervical lesions,and to design a new algorithm based on the color and texture information of the female cervix in the original image.It is used to realize the extraction of the region of interest(ROI)of the cervix and the location and segmentation of the lesions position.Taking the dichotomous classification of cervical lesions as an example,through multiple comparison experiments,the accuracy of the final cervical lesion classification prediction was raised up to 92.95%;compared with the unimproved method,the time complexity and space complexity are significantly reduced.Experiments have proved that the proposed method is feasible,which can assist radiologists in the diagnosis of cancer levels,and improve the efficiency and accuracy of diagnosis. |
