Research On Medical Image Recognition Algorithm Based On Convolutional Neural Network

With the development of medical imaging technology,the use of medical imaging for disease diagnosis has become the main diagnostic method in major hospitals.The medical imaging diagnosis requires experienced doctors,but there are few of them,and the diagnosis of a large number of medical images only by manual diagnosis may lead to fatigue diagnosis and even misdiagnosis.With the development and application of deep learning technology,the application of artificial intelligence in medical image recognition has attracted academic attention.Due to the special imaging form of medical image data,the small number of datasets and the difficulty in extracting image features,the current algorithms have low accuracy and high time complexity for medical image recognition.In order to improve the accuracy of medical image recognition,two-dimensional dermoscopic images of skin melanoma and threedimensional brain magnetic resonance images of Alzheimer’s disease were used as examples.The main research contents are as follows:(1)An Inception Deep Residual Network(IDRN)algorithm is proposed to classify cutaneous melanoma.In this algorithm,Inception structure was used to replace the convolutional pooling layer in the Residual Network(Res Net),and SELU activation function was used in the whole Network.Using the ISIC2017 skin melanoma dataset as the training set of the model.Theory and experiments show that the proposed IDRN classification algorithm reduces the time complexity and improves the recognition accuracy.(2)The Xception Deep Residual Network(XDRN)is proposed to identify and classify the slice data of brain MRI.Xception network structure is introduced to replace Inception structure in IDRN algorithm and introduced a global pooling layer.The ADNI brain MRI dataset is sliced and the slices with large image information are selected as the training data set of the model according to the image entropy in the obtained slice data.The experimental results show that the proposed XDRN algorithm has improved the accuracy in the recognition and classification task of Alzheimer’s disease..(3)An integrated learning algorithm for Alzheimer’s disease based on adaptive weighted fusion is proposed.By forming three different data sets from the sagittal plane,coronal plane and cross-section plane of the ADNI dataset,three base classifiers are formed by XDRN algorithm training and the results of the three base classifiers were weighted and fused by ensemble learning algorithm.Experiments show that the recognition accuracy of the proposed integrated learning model is further improved than that of XDRN algorithm.