Research On Medical Image Detection And Segmentation Algorithm Based On Deep Learning

Medical images play a crucial role in the medical field,providing objective evidence for doctors and helping diagnose patients’ conditions.With the development of medical imaging technology,the number of medical images has increased dramatically,and traditional manual processing methods can no longer meet the image analysis needs of modern medicine.Therefore,deep learning,as an emerging artificial intelligence technology,can automatically analyze image features and improve the efficiency of medical image processing,and is receiving increasing attention.In deep learning-based medical image processing,detection and segmentation are two key tasks that are responsible for locating abnormal areas and separating lesion sites,providing more accurate diagnostic information for doctors.Unlike natural images,medical images have highly complex characteristics such as significant interference and noise,diverse scales and shapes of lesion tissues,and low contrast,which make general image detection and segmentation algorithms often unable to handle complex medical image processing tasks.Aiming at the characteristics of medical images,a novel image detection algorithm based on attention mechanism and an image segmentation algorithm based on multi-scale semantic representation are proposed in this paper.Furthermore,an intelligent medical image processing system is constructed based on the above algorithms.The main work is as follows :(1)To address the problem of significant interference caused by a large amount of noise and irrelevant content in medical image detection tasks,a medical image detection algorithm called FCOS-plus based on attention mechanism is proposed in this paper.The algorithm is built on the architecture of the FCOS algorithm and uses the Swin Transformer as the backbone network.It efficiently mines the representation information of the input image through the multi-head self-attention mechanism and establishes cross-window information exchange through the shifted window method,thus fully extracting the global and local information of the features.At the same time,a feature pyramid with integrated channel attention mechanism is designed to increase the weight of key regions in the image and suppress irrelevant information in other regions.In addition,deformable convolution is introduced at the end of the algorithm to enhance the model’s ability to model unknown target shapes.Experiments on a tuberculosis bacterial detection dataset show that the average precision of the improved FCOS-plus algorithm reaches 85.7%,and the recall rate reaches 97.8%,which is 4.9% and 2.8% higher than the original FCOS algorithm,respectively.This indicates that the FCOS-plus algorithm has better performance in medical image detection.(2)To address the problem of complex tissue morphology,diverse scale and geometry shape in medical image segmentation tasks,a medical image segmentation algorithm based on multi-scale semantic representation is proposed in this paper.Firstly,Res2 Net Blocks are used to extract image features,and the group residual connection structure is used to express the multi-scale features of the image at a finer granularity level.Then,to fully utilize these features,the spatial pyramid pooling module and feature pyramid are used to fully fuse information of different scales and solve the problem of detail loss during segmentation.Finally,the Dice Loss is used as the main loss function to calculate the difference between the predicted values of the network and the ground truth labels,achieving pixel-level classification of medical images.The experimental results show that the F1-Score of the proposed algorithm on the retinal vessel segmentation datasets STARE and CHASE＿DB1 are 0.8606 and 0.8760 respectively,and the sensitivity are 0.8544 and0.8745 respectively.This is better than commonly used semantic segmentation algorithms such as U-Net,R2 U-Net,and DUNet,indicating that the proposed multi-scale algorithm has good segmentation performance.(3)A medical image intelligent processing system is designed and developed using the Model-View-ViewModel software architecture pattern in this paper,based on the two aforementioned algorithms.To meet the needs of medical professionals,mainstream development technologies such as Vue,Express,and MySQL are used to implement system functions such as patient management,intelligent detection of tuberculosis bacteria,automatic segmentation of retinal blood vessels,and inspection report management.The system’s usability is validated through functional and compatibility testing,and the results show that the system can effectively assist doctors in processing medical images.