Design And Application Of Chest Medical Imaging Auxiliary Diagnosis Algorithms Based On Deep Learning

In recent years,the rapid development of artificial intelligence theory and practice rep-resented by deep learning algorithms has promoted changes in many research and applica-tion fields.Deep convolutional neural networks have been widely used in major computer applications.In some tasks they have even exceeded human recognition capabilities.In re-sponse to the government's "intelligent medical" construction concept,auxiliary diagnosis algorithms and application research that intersect with medicine have became a hot topic in the field of computer vision.Many medical imaging auxiliary diagnosis algorithms based on deep learning have emerged.Medical imaging auxiliary diagnosis system based on deep learning algorithms can effectively alleviate the workload of clinical radiologists and reduce the occurrence of missed diagnosis or even misdiagnosis.This paper studies the design and application of thoracic medical imaging auxiliary diagnosis algorithms based on deep learning,summarizes the previous research work,and proposes improvements.The main work includes the following three aspects:(1)To overcome the limitation of object detection algorithm in the task of rib fracture detec-tion,we propose a rib fracture detection algorithm of spatial multi-scale network fusion based on the lightweight MobileNet network.The rib fracture detection is divided into two stages:first,the skeletal area in the medical image is extracted based on the graph-ics algorithm,and rib images of multiple scales are obtained by linear interpolation;then the spatial multi-scale 3D MobileNet convolutional neural network model fusion is used to realize the classification of rib fracture.The experimental results on the image dataset from a hospital show that the fusion of spatial multi-scale models can effectively improve the accuracy of rib fracture detection.(2)In order to improve the accuracy of convolutional neural network nodule detection and reduce false positives,we propose a pulmonary nodule detection algorithm based on the attention mechanism.The 3D convolutional neural network with the attention mech-anism enhances the feature maps at the channel and spatial levels,respectively.The output layer of the 3D region proposal network is used to generate candidate pulmonary nodule regions.Finally,the non-maximum suppression algorithm is used to remove redundant candidate nodule regions to obtain the pulmonary nodule detection results.The experimental results on the LUNA 16 dataset show that the convolutional neural network with attention mechanism can effectively improve the accuracy of pulmonary nodule detection.(3)Combining the two major chest medical image detection tasks in(1)and(2),a deep learning-based chest medical imaging auxiliary diagnosis system is designed and imple-mented.The auxiliary diagnosis system is divided into three modules:a data persistence module,a model inference module,and a diagnosis visualization module.According to the type of examination of the patient,the auxiliary diagnosis system conducts the cor-responding data pre-processing and model inference through the processing pipeline in the background,and visualizes the test results as diagnosis suggestions to the clinical radiologists to assist their diagnosis work.