Research On Recognition And Segmentation Of Osteoporosis MRI Image Based On Deep Learning

Osteoporosis is a systemic bone disease,which is characterized by damage to the bone microstructure and decreased bone density,which in turn leads to increased bone fragility.Especially with the aging of the population,osteoporosis has become an epidemic that seriously threatens the health of middle-aged and elderly people.At present,the clinical diagnosis of osteoporosis is confirmed by dual-energy X-ray bone densitometer(DXA).DXA is mainly based on bone density.The clinical diagnosis of osteoporosis mainly depends on the experience of doctors,and although some DXA shows Low bone mass images are judged to be non-osteoporotic,so the diagnosis may be misjudged due to subjectivity.Therefore,the development of related computer-aided diagnosis systems to assist in the diagnosis of osteoporosis has very important practical significance for clinical assisted diagnosis.This paper mainly uses deep learning algorithms to conduct in-depth research on MRI images of osteoporosis,which mainly includes two aspects: diagnosis classification and image segmentation.First of all,in view of the current clinical diagnosis of osteoporosis,which mainly depends on professional physicians and the value of bone density,this paper proposes a CNN-HKNN deep learning model.In the CNN-HKNN model,the CNN network extracts the texture features of the image through convolution and pooling,and uses the improved HKNN algorithm to replace the traditional Soft Max algorithm as the model classifier.And this model uses Gabor filtering to perform multi-scale and multi-directional filtering processing on osteoporotic MRI images,which enhances the texture features of the image and expands the MRI data set.Secondly,for the spine bone segmentation of osteoporotic MRI images,this paper proposes a Unet-CRFs segmentation method,which uses the U-Net network to extract the spine bones in MRI,and combines the fully connected condition random field(Dense CRFs)to improve the edge details of the spine bones.In addition,in view of the quality of osteoporotic MRI images,after cropping and normalizing the images,this paper uses the image spatial frequency and appropriate thresholds to quantify the quality evaluation indicators of osteoporotic images.This effectively reduces the noise caused by poor quality images.We validated the research in this article with the collected osteoporosis data set.For the CNN-HKNN model,the osteoporosis MRI data set is classified and identified by a 10-fold cross-validation method.Experiments show that the accuracy of the model proposed in this paper is 96.3%.This model can be useful for the clinical diagnosis of osteoporosis.Supporting role.At the same time,the Unet-CRFs segmentation model in this article was verified.Experiments show that this model can segment the spine bones of MRI images better in the osteoporosis MRI data set,and has better results than a single U-Net segmentation.In summary,based on deep learning,this paper conducts research on the recognition and segmentation of osteoporosis MRI images,and proposes a clinical auxiliary diagnosis model for osteoporosis,which can achieve high recognition accuracy in osteoporosis MRI images.At the same time,this article applies the Unet-CRFs network model to the spine bone segmentation of osteoporotic MRI and achieves a better segmentation effect.It is of great significance in assisting the clinical differential diagnosis of osteoporosis.