Risk Classification Of Early Visceral Cancer Based On Facial Images

The number of new cases of cancer has been increasing in recent years.Clinically,cancer can be divided into five stages according to the extent of its spread,in which a certain part of the cancer cells for stage I,in the cancer site for the slight spread of Stage II,these two periods are collectively referred to as the early stage of cancer.Early cancer patients can be cured by means of chemotherapy and radiotherapy.Therefore,early detection of cancer patients and their families is of great significance.As a traditional Chinese medicine,traditional Chinese medicine has been handed down in China for thousands of years.Doctors of traditional Chinese Medicine(TCM)diagnose patients’ conditions by observing,asking,examining and consulting.As the main content of inspection technique,face-to-face diagnosis plays an important role in diagnosis.In recent years,the development of information technology of traditional Chinese medicine is becoming more and more popular.The combination of traditional Chinese medicine and information technology not only promotes the development and dissemination of traditional Chinese Medicine Technology,but also avoids the influence of subjective experience of doctors to some extent.Based on face-to-face diagnosis technology of traditional Chinese medicine(TCM).This thesis is based on the study of facial color and morphology,and uses the random forest algorithm to classify whether there is a risk of early visceral cancer,and based on the VGG16 network structure and migration learning,which specific visceral cancer risk classification.1: The improved Bise Net algorithm is used to segment the face and tongueimages.In the process of image acquisition,the same equipment is used to ensure the consistency and stability of the light source.But there will inevitably be other background noise effects.In order to get a more accurate model,we use the improved Bise Net algorithm to segment the acquired image,obtain the facial image and remove the background noise.In the research,pyramid structure is introduced into Bise Net network to make the segmentation result more accurate.2: The YCBCR Color Space has the characteristic of ellipse skin colorclustering,according to this characteristic,skin color points and non-skin color points can be distinguished,and the non-skin color points can be denoised by 9 * 9 mean filter,andthe brightness(Y)and blue component(CB)can be calculated by color moments,the mean value of the red component(CR)is used as its color feature.The method of gray level co-occurrence Matrix is used to calculate the ASM energy value,entropy and contrast of the image as its texture features and reflect the facial texture information.Then a random forest classification model was constructed.The random forest is composed of many decision trees,which is constructed by ID3 algorithm.By adjusting the maximum feature number of the decision tree and the number of the decision tree in the random forest,the optimal parameter model is found,which can achieve the optimal result with the least time consuming.3: Because of the small number of samples in this research,when training with deep learning network,the network parameters are more,and the model is over-fitting,which leads to poor performance and low classification accuracy.To solve this problem,this study was based on VGG16 and migration learning to classify the types of visceral cancer.For the VGG16 network model trained on Image Net data set,the transfer learning method is used to transfer the model to the research of the classification of visceral cancer.From the point of view of face-to-face diagnosis of traditional Chinese medicine,through the features of facial images,the risk of visceral cancer can be predicted.In the machine learning classification research and depth learning classification research to explore the comparison,the final realization of a fast,high-accuracy classification with or without visceral cancer risk.