Digital Histopathology Image Classification Algorithm Based On Convolutional Neural Network

Pathological diagnosis is the most direct and accurate gold standard in the clinical diagnosis of cancer.Traditional pathologists’ method of reading films under artificial microscope is not only susceptible to subjective experience,but also time-consuming and laborious due to the amount of pathological image data.With the maturity of digital pathological imaging technology,panoramic digital pathology has gradually become an important tool for clinicopathological diagnosis.However,panoramic digital histopathological images are characterized by high resolution,large scale and complexity,so it is still a technical challenge to realize automatic processing and intelligent analysis of pathological section images.Traditional machine learning algorithms,which rely on manual extraction and analysis of features,are very difficult and consume a lot of manpower and material resources in processing super large pixel pathological images,which cannot meet the needs of actual clinical application.Although the algorithm of convolutional neural network improves the automatic feature extraction and computational efficiency,and is widely used in medical image analysis,the problem of how to overcome the computational analysis of panoramic pathology still exists.In order to solve the above problems,this paper conducts in depth research and verification on the histopathological image processing by using the method of convolutional neural network.From the clinical practice design more focused efficiency and explanatory classification algorithm;The clinical application scenarios and value of convolutional neural network were analyzed and verified.Based on the characteristics of pathological images such as large size,rich information and hierarchical storage,this paper firstly proposes a fast method of detecting the region of interest of pathological images based on convolutional neural network.Different from the method of direct segmentation,which requires more precise contour labels,the method of this paper can complete the training of the model only with the category label of the image,which is simpler and faster than the method of direct segmentation.Secondly,in terms of classification algorithm,this paper proposes a classification method of deep multi scale convolutional neural network based on attention mechanism by combining the steps of clinical diagnosis and related theories,which is different from the traditional method of using single image scale for pathological image analysis.Methods by learning different weight ratios of different scale features,and then learning the features of multiple scales by fusion,the classification is finally realized.The result statistics of the algorithm also proves the feasibility of this method.Finally,in the actual classification diagnosis,this paper applied the method of convolutional neural network to study the classification of clinical histophological images,and carried out in depth comparative analysis between the algorithm results and pathologists.The average accuracy of the convolutional network model was 0.905.