Key Techniques Of Radiofrequency Ablation Surgery Planning For Liver Tumors

Liver cancer is one of the most common malignant tumors,and it takes the life of more than one million patients every year worldwide.Radiofrequency ablation(RFA)has been demonstrated to be an important technique for minimally invasive interventional treatment of liver cancer.It is the preferred therapy for unresectable cancers when the tumor is too large,the tumor is too close to a large blood vessel or the tumor has spread throughout the liver organ.In traditional RFA surgery for liver cancer,surgeons roughly determine the required number of ablation and needle insertion positions according to patients' CT images and their clinical experiences.There is a lack of quantitative planning methods to accurately calculate and plan the ablation region,resulting in excessive ablation(damage to normal tissue),or incomplete tumor ablation,especially for large tumors.Surgeons usually insert ablation needles under the guidance of ultrasound,CT or C-arm images during surgery.The insertion process highly relies on surgeons' experiences,and is affected by human factors such as spatial imagination Inaccurate needle placement often happens during treatment.With the development of medical robots and navigation technology,robot assisted RFA surgery is studied to achieve ablation surgery with high accuracy.RFA surgery planning for liver tumors is an important part of robot assisted RFA surgeryTaking robot-assisted RFA surgery as background and aiming at clinical needs in RFA surgery,this thesis studied the key technologies in RFA surgery planning for liver tumors,including accurate liver and liver tumor segmentation and RFA insertion path planning.Based on these technologies,we established patient-specific three-dimensional models,and obtained accurate RFA insertion paths to ensure the thermal ablation area completely cover the tumor in three-dimensional space,and not damage the key structures such as large liver vessels,thus improve the planning accuracy of RFA surgeryThe contribution of this thesis can be summarized as follows:(1)We proposed a 2D/3D CNN hybrid liver tumor segmentation method DFN+ DS-3D-UNet-ori,which was superior to all 3D CNN-based methods and most 2D CNN-based methods on the LiTS 2017 test data set.(2)We presented DS-SE-V-Net by introduction of deep supervision and Squeeze&excitation modules to original V-Net,to automatically segment brain tumors from multi-model MR images,and obtained better results than traditional encoder-decoder models on the BraTS 2017 dataset.(3)We proposed an interactive RFA surgery planning method based on geometry constraint for single insertion port RFA surgery and a hexagonal close-packed sphere propagation algorithm for overlapping ablation planning of large tumors,which could obtain conformal coverage of arbitrarily shaped large liver tumors with as few ablation times as possible.Finally,several phantom experiments and two animal experiments with pigs were carried out by the image segmentation software and interactive RFA surgery planning software developed in this thesis and the RFA robot developed by the project team.Experiments proved the robot assisted RFA system could achieve multi-needle insertion stably and accurately through the single insertion port(SIP).This system greatly reduces the reliance on surgeons' experience and the invasiveness of RFA surgery,and is also beneficial to the standardization of RFA surgery for large tumors.