Research On Medical Image Processing Driven By Multi Task Clustering Algorithm

With the development of artificial intelligence,smart medical care has become a hot smart application field,and it plays an important role in medical image processing.However,with the continuous development of these technologies in the field of clinical medicine,people’s dependence on images has gradually increased,and the demand for diagnostic accuracy has continued to increase.Therefore,medical image processing technology still has great room for improvement and development.Among them,image segmentation has always been a difficult and breakthrough point in this field,especially for images with uneven grayscale,partial volume effects,blurry and complex images such as brain MRI.It is difficult to obtain high accuracy.Because of the differences in the internal tissues of each human brain,in the process of analyzing MR brain images,accurately obtaining the location and volume of white matter,gray matter,and cerebrospinal fluid in the brain image is an important step for subsequent medical diagnosis.Therefore,this article is based on multi-task clustering algorithm to realize the segmentation of brain images,thereby improving the anti-noise ability in the segmentation process and obtaining higher segmentation accuracy.The specific work in the research process of this subject is as follows:(1)The segmentation of brain MRI images is mainly realized by clustering machine learning method.Traditional clustering methods are single task clustering.Single task learning can only mine limited information from a single object,and can not effectively use the association between tasks to get more useful information,so it is limited in dealing with multi task scenes.This paper designs a new multi task collaborative learning framework,which is combined with fuzzy system to improve its learning ability and interpretability.(2)A new multi task weighted fuzzy C-means(MT-WFCM)clustering algorithm is proposed,which shares the common clustering centroid representing the relevant information of different tasks,and avoids the negative impact of noise in the original data instances.At the same time,multiple task data are processed to extract the common information between tasks,and the clustering performance of each task is improved by using this information.At the same time,a simple sampling strategy is adopted in the distributed mt-fcm segmentation algorithm,which reduces the clustering time consumption and is conducive to the practicability of the algorithm,so as to improve the segmentation effect and complete the segmentation task.(3)A new multitask quadratic regularized clustering algorithm(MT-QRC)is proposed for magnetic resonance brain image segmentation.Although the fuzzy c-means algorithm can well describe the fuzzy situation of MR image in image processing,it depends on the selection of initial parameters such as initial clustering center,and its improvement method is relatively simple.Therefore,we try to choose the square entropy algorithm as the basic model and replace the fuzzy index in fuzzy clustering with the parameter term controlled by square entropy,The limitation of fuzzy index in clustering is reduced and the flexibility of parameters is enhanced.It is also combined with the multi task learning framework to transfer the relationship between various tasks.(4)An improved multi view fuzzy C-means(IMV-FCM)is proposed by using the view weight adaptive learning mechanism,so that each view can obtain the best weight according to its clustering contribution.Finally,the segmentation result is obtained by the view integration method.Under the view weight adaptive learning mechanism,the coordination between views is more flexible,and each view can learn adaptively to achieve better clustering effect.(5)The experiments of multi-task and multi-view clustering algorithm proposed in this paper show that it has good segmentation effect of brain MRI image and improves efficiency and accuracy.At the same time,according to the importance of different perspectives,the corresponding weights are adjusted to obtain better performance in MR image segmentation.