Empirical modeling of renal motion for improved targeting during focused ultrasound surgery

Non-invasive surgery looks at ways of eliminating physical contact with the target tissues while maintaining necessary levels of accuracy. Focused Ultrasound Surgery (FUS) is one such treatment modality, which uses a tightly focused beam of high intensity ultrasound to ablate tumors in various parts of the body. For trans-abdominal access, respiration induced movement of the tissue targets remains a major issue during FUS. Respiration induced movements are known to be significant in liver and kidney. In this paper, we attempt to address this problem using non-linear prediction and modeling techniques as applicable to kidney movement patterns. Kidney movement patterns are known to be three dimensional and vastly complicated compared to movement patterns of the liver. Monitoring and quantification of the nature and extent of kidney movement is yet to be explored in depth for effective compensation and accurate targeting. Apart from the respiratory cycle, the movement of the kidney is also affected by several factors, such as the movement of the ribs, spleen and liver. Modeling of these movements is imperative for motion adaptive FUS. Since kidney movements are highly subject specific, generic statistical models cannot be used for compensation. The system latency and real-time performance of the imaging modality also induce additional parametric dependence in target tracking. In this work, we focus on empirical modeling and prediction of the kidney movement to for error analysis and computing system latency. The accuracy of existing modeling techniques is compared with a newly developed empirical model. From the study conducted in healthy volunteers, it was found that the kidney movement was complex and subject specific and could be effectively modeled using the new shape function based model. The model was further fine-tuned using Kalman filter based predictors and Adaptive Neuro-Fuzzy Inference System (ANFIS) which gave more than 85% accuracy in prediction.