Research On In-situ Skin 3D Printing Navigation Technology Based On Deep Learning

In recent years,biological 3D printing has promoted the rapid development of techniques such as tissue engineering and drug screening,enabling the in vitro construction of tissues and organs.The rise of in-situ skin 3D printing technology provides new techniques for biomimetic and accurate restoration of defective skin,and automatic in-situ skin 3D printing has become a trend in the future.Fully automatic in-situ skin 3D printing requires precise detection of the location of the defected skin,judging the type of damage,and then achieving full-automatic integrated resection,printing,and repair.Therefore,automated navigation technology is a crucial step in fully automatic in-skin 3D printing.To automate in-situ skin 3D printing,you first need to accurately segment the defective skin area.However,there are relatively few researches on the segmentation of defective skin in in-skin 3D printing.The classical medical image segmentation algorithm has poor performance in segmentation of skin image noise and hair and so on,so it is difficult to solve the defect skin segmentation faced in in-skin 3D printing.The actual engineering problem.In recent years,with the development of deep learning,it has achieved better results than traditional methods in image segmentation.In order to overcome the shortcomings of traditional segmentation algorithms in the segmentation of defective skin images,this paper proposes a Dense U-Net-based defect skin segmentation algorithm for the first time and applies it to the in-situ skin 3D printing navigation technology task,which improves the accuracy of image segmentation and enhances The algorithm anti-jamming capability.1)First,research on denoising and data enhancement of deficient skin images was performed.The defect skin image in the in-situ skin 3D printing scene is usually affected by noises such as hair,black frames,bubbles,etc.A method of denoising a defective skin image is proposed.In the project implementation of the deficient skin segmentation,it is difficult to obtain a large number of defective skin images and label them at the pixel level.Therefore,the research focus of this scheme is on the research and implementation of the data enhancement of the decomposed skin segmentation data set.2)Secondly,this paper proposes a method based on Dense U-Net to solve the defect skin segmentation problem.The algorithm achieves end-to-end segmentation by merging dense convolution and full convolutional neural networks.In order to further improve the segmentation results,a new activation function was introduced and various measures to prevent over-fitting were adopted.The training stage combined the transfer learning with the two-stage training.Experimental results show that the algorithm has deeper network layers,smaller parameter quantities,higher segmentation accuracy,and better anti-overfitting performance.3)Finally,the segmentation algorithm is ported to the in-situ skin 3D print navigation system.The software environment and implementation process of this algorithm are introduced.The realization and results of the segmentation algorithm are analyzed.By comparing the experimental results,compared with the traditional defect skin segmentation algorithm,this algorithm has high accuracy in segmentation detection,better real-time performance,good robustness against noise interference such as hair,and good engineering application value.This article has solved the contour recognition problem of the defect skin in the automatic navigation of in-situ skin 3D printing,and can measure the shape,boundary,etc.of the defect skin,and can help the doctor to formulate and modify the treatment plan and achieve full automatic in-situ skin 3D printing.