Research And Implementation Of Localization Method For Focal Leather Dysplasia Based On Deep Learning

The rapid development of artificial intelligence and medical imaging science in their respective fields has led more and more countries to promote it as a national strategy to promote the product revolution and social revolution.The amount of medical imaging data is large,and professional interpretation often takes a lot of time of experts.At the same time,the learning of professional knowledge and accumulation of clinical experience are also indispensable.In recent years,deep learning technology has achieved great success in many aspects of medical imaging science.The application of artificial intelligence in the medical field has now become one of the hot research fields.As one of the intractable problems in the medical field,focal cortical dysplasia(FCD)is one of the subtypes of cerebral cortical dysplasia and is a common cause of drug-refractory epilepsy.Viral infection and somatic mutations may be involved in the pathogenesis,but the specific mechanism is still not clear.With the development of functional neuroimaging technology,FCD has been proven to be one of the main reasons for surgical treatment of patients with drug-refractory epilepsy.Resection of the lesion can bring good results for most patients.This study proposes a deep learning-based FCD lesion localization method.By effectively processing and analyzing three-dimensional magnetic resonance imaging data,it can locate the lesion,assist doctors in diagnosis and treatment,and provide professional scientific basis for patient treatment.This paper first focuses on learning and research on the relevant network framework of full convolutional neural networks in deep learning,and proposes solutions for FCD lesion localization in two and three directions.Based on the clinical experience of experts,the original 3D image is preprocessed,including skull peeling,feature map extraction and other processes,and then the lesion segmentation scene for medical images is under the FCN and U-Net(3D U-Net)network framework Design and train network models,and evaluate the models by comparing the evaluation indexes of the two models.In addition,considering that the FCD lesions have different shapes and sizes in the brains of different patients,a long-range dependency acquisition module is proposed based on this module,and an Ld-Net network is designed based on this module.3D Ld-Net in Ld Based on-Net,further improvements have been made to improve the positioning of the lesion.This paper confirms that the idea of applying the full convolutional neural network framework structure in deep learning to the location of FCD lesions is feasible from two and three perspectives.Based on the characteristics of the lesion,a long-range dependency acquisition module was proposed to further improve the model positioning effect.In addition,it also utilizes medical image related data preprocessing technology to complete the entire method system from data preprocessing to lesion detection.The results provide a reference for FCD lesion localization research based on medical images.