Research On Image Segmentation Algorithm Of Laser Ablation Of Soft Tissue

With the rapid development of the current medical treatment level,many treatment plans for minimally invasive surgery are derived.These plans are based on diverse medical imaging technologies and efficient computer image processing.Laser interstitial thermal therapy(LITT)is a treatment plan combined with laser ablation equipment and effective thermal ablation zone control.Because the heat transfer equation of the human tissue(Pennes equation)is the gold standard equation for thermal area simulation,but it can only simulate the thermal area,but it cannot cooperate with MRI,CT and other medical imaging equipment for timely feedback;The minimally invasive surgery of organs has extremely high requirements on the control accuracy of the instrument.The Pennes equation cannot meet the millimeter-level simulation experiments.Therefore,there are few programs for the study of laser ablation thermal regions through image quantization.In recent years,the effective control of the thermal ablation area is mainly based on image data,and the image segmentation technology can provide effective information feedback for doctors in a short time,greatly improving the efficiency of medicine during surgery.Medical image segmentation is technically an effective extraction of the target area.The extracted image information is combined with the doctor's prior knowledge.The computer can further analyze the graphic information to obtain effective visual images or data information.The form of graphics plus text provides doctors with favorable surgical guidance,but the current image segmentation is difficult to accurately mark,the segmentation accuracy cannot be controlled below the millimeter level,and the algorithm parameters are numerous.In view of the above problems,this paper researches and improves traditional medical graphics algorithms and image segmentation algorithms under deep learning.The aim is to use a suitable algorithm for the segmentation of laser ablation graphics,so as to achieve the quantization ability of millimeter-level image data.First,this paper analyzes classic and popular algorithms in recent years,including clustering segmentation algorithms,optimization-based threshold segmentation,active contourbased segmentation,atlas-based segmentation,etc.,combined with the problems of laser ablation images in this paper,such as difficult to segment and difficult to mark,The k-means algorithm,the mean-shift algorithm and the level set algorithm in the clustering algorithm are selected for analysis and improvement.Ms))algorithm to cooperate with the level set algorithm for automatic parametric image segmentation,and the success of the algorithm improvement is verified by multi-modal images.Secondly,the U-net model in the deep neural network is learned,and the U-net algorithm is improved in two aspects,including multi-scale convolution kernel,dense network and other optimization schemes,and the IDU-net algorithm is proposed in the feature extraction process.Get more content in,and can further reduce the network parameters to speed up the calculation speed,and through the brain MRI images to verify the optimized design of this article,the accuracy is increased by 12%.Finally,this paper also builds an experimental environment through an independently designed laser ablation experiment,a 980nm laser generator,a dispersion fiber,and a condensation device,and takes pig liver as the ablation target,and obtains the corresponding laser ablation image through frozen section and image acquisition.Choose the self-adjusting k-means algorithm in the traditional algorithm,the RSF level set segmentation algorithm based on self-adjusting mean shift,and the IDU-net algorithm to segment and quantize the final laser ablation experiment image,which effectively segmented the image,Quantifying image data also confirmed the effectiveness of the three algorithms,and the quantization error is less than 10%.