| Cancer is one of the most common diseases that cause human death,and it is also an important factor restricting the rapid development of human society and civilization.In recent years,the incidence of cancer has shown an increasing trend,but the imbalance between the number of professional doctors and patients has led to the low efficiency of cancer and precancerous lesions screening.Research on efficient and accurate image recognition algorithm is one of the important means of intelligent cancer screening,which can effectively reduce the workload of doctors and improve the screening efficiency of cancer lesions.Convolutional neural network is often used in the design of image recognition algorithms with the advantages of strong antiinterference ability and good migration ability.However,in the face of the high similarity between various types of cancer image lesions,the recognition ability of existing models is still insufficient.This paper will build a cancer image lesion recognition algorithm based on convolutional neural network,and improve the performance of the network in cancer image lesion recognition from the perspective of feature optimization.Based on this algorithm,an intelligent cancer auxiliary diagnosis system is designed to provide technical support for the recognition of lesions in cancer images.The main contents of this paper are as follows:(1)An image classification method combining Gabor texture enhancement and convolutional neural network is proposed.Firstly,the Gabor modulation convolution module is designed by exploiting the sensitivity of Gabor wavelet transform in direction and scale.In this module,Gabor filters are used to modulate the convolution kernel parameters,thereby restricting the optimization direction of network parameters,and information aggregation is carried out through channel Max pooling to improve the convergence speed and recognition ability of the network.Then,the convolutional structure at the end of each stage in Res Net50 was replaced by this module,and the GCFM-Res Net50 model with texture enhancement performance was proposed.Finally,the model was tested on the HAM10000 skin cancer classification dataset.The results show that the proposed model improves the average class accuracy and F1 score by3.26%and 2.57% respectively compared with the baseline model,and it also has advantages compared with other excellent models proposed in recent years,indicating the effectiveness of the algorithm.(2)An object detection method based on multi-scale detail enhanced pyramid is proposed.The multi-scale detail enhancement pyramid combines the Gabor modulation convolution model with the feature pyramid network model to extract different scale features from each stage of the feature extraction network,and the texture information extraction ability of the Gabor modulation module is used to enhance the features of each scale to avoid the problem of information loss in multi-scale feature fusion.In addition,a dual-channel pooling module is designed to aggregate spatial information from two dimensions of direction and channel,which retains the similarity in direction and the correlation in local position of features at each stage,and eliminates the information difference during feature fusion.This module is used to improve the Sparse R-CNN model and test it on the self-built esophageal cancer detection dataset.The results show that the algorithm improves the commonly used m AP.50 index by 2.4 % compared with the baseline model,and also obtains the highest detection accuracy compared with other existing models.(3)An AI-assisted cancer screening platform is designed and constructed.The AI-assisted cancer screening platform aims at the early screening of skin cancer and esophageal cancer.Based on the skin cancer recognition algorithm and esophageal cancer detection algorithm proposed in this paper,the skin cancer and esophageal cancer detection interfaces are designed respectively,which can realize the automatic recognition of two kinds of cancer images through the form of front-end uploading images and back-end intelligent analysis.The platform integrates intelligence,interactive,simple operation,can effectively assist doctors to diagnose lesions,improve diagnosis efficiency,and has certain application value. |
