Prediction Of Soft Tissue Thickness Based On Machine Learning And Research On Craniofacial Restoration

In the field of skull face restoration,the acquisition of soft tissue thickness of facial feature points is a very important step.At present,the acquisition of soft tissue thickness at facial feature points of skull is mainly achieved by the cooperation of advanced medical equipment and professional medical staff,but there is a great defect in this method,that is,the thickness measurement of the face can not be carried out for the skeleton.In recent years,the average method is widely used,that is,the average thickness of soft tissue at the facial feature point sampled by a large number of samples is used as the thickness of soft tissue at the facial feature point of the skull to produce the skeleton appearance,but the difference between bones has not been taken into account,which makes the appearance of restoration similar.In this paper,the deep learning method is used to study the relationship and features between the skull and the face through the nonlinear mapping of the hidden layer in the network,so as to predict the thickness of the soft tissue of an unknown person's face according to the characteristics of the remains themselves,and then solve the problem of directly predicting the facial shape according to the skull.This thesis is divided into two parts:The prediction of facial soft tissue thickness based on feedback neural network is proposed.In the case of giving the prior term of mean soft tissue thickness at facial feature points,we can better solve the difficult problem of predicting soft tissue thickness at facial feature points of bones by learning the mapping relationship between the skull surface feature points of adults and the corresponding three-dimensional spatial coordinates of skull facial feature points^[27].The model takes the 3d spatial coordinates at the feature points of the skull and face surface as the input and output of the model.Experimental results show that the predicted error mean of the proposed method is that the soft tissue thickness at 35 facial feature points is less than 1 mm,meet the clinical needs.Furthermore,a comparative analysis of the proposed method and Dinh et al shows that the proposed method is more accurate in predicting soft tissue thickness at facial feature points.The prediction of soft tissue thickness from the end to the end of the end face of the deep convolution codec based on the hopping layer connection is presented.According to the principle of cylindrical projection,the three-dimensional mapping position two-dimensional elevation map data are obtained.The generated two-dimensional skull and face elevation map is taken as the input and output of the network,and then the predicted facial elevation map is analyzed by the method of back projection visualization and two-dimensional and three-dimensional thermal map.At the same time,the soft tissue thickness predicted by this method is compared with the prediction result of BP feedback neural network and the latest prediction result data.It is verified that the deep convolution codec is used to predict the soft tissue thickness,which fully considers the spatial characteristics of facial points,makes up for the errors caused by sparse feature points,and makes the prediction results more accurate.Moreover,the optimal training model is selected by comparing and analyzing between different networks and between different prediction sizes through two evaluation indexes of peak signal-to-noise ratio(PSNR)and structural similarity(SSIM).