MR image-based automatic control of high-intensity ultrasound thermal therapies of cancer.

摘要

Thermal therapies, which involve the use of elevated temperatures to selectively heat the target, have shown promise as noninvasive medical interventions for tumor treatment. Long treatment times, incomplete treatment of large targets, and unintended normal tissue damage continue to impede a broader penetration of high-intensity focused ultrasound (HIFU) therapies into clinical practice. Planning and control of noninvasive thermal therapies can be further improved if site- and patient-specific thermal and power deposition models of the treatment are available.; Model-based automatic treatment control system has the potential to address these control related issues systematically. Two different control strategies were developed to achieve therapeutical goals of delivering physician prescribed thermal dose to the target in minimum treatment time without violating explicitly-imposed normal tissue safety constraints. In particular, a constrained model predictive controller (MPC) was developed to control the delivery of the thermal dose, while satisfying normal tissue temperature constraints in order to guarantee treatment safety. Another treatment control system was developed based on the necessary conditions of time-optimality. The treatment control systems automatically select the focal zone trajectory and the intensity of the applied ultrasound transducer to balance the clinically conflict efficacy and safety objectives. The controllers are characterized by their time-optimal performance and direct incorporation of normal tissue constraints. The developed approaches were evaluated in computer simulated treatment of three-dimensional patient models heated with ultrasound phased array system.; Since the developed treatment control systems are model based and use MRI feedback, it was necessary to develop novel methods to identify reduced-order patient- and site-specific treatment models based on massive amount of data provided by real-time MR thermometry images. An image-based approach to the dynamic identification of low-dimensional, patient- and site-specific noninvasive thermal treatment models using proper orthogonal decomposition (POD) of MR thermometry images has been developed. The developed methods are less sensitive to temporal and spatial noises in image data, slow image acquisition rate, and are suitable for adaptive model re-identification by recursively utilizing newly acquired images. Three-dimensional computer simulations and MRI thermometry experiments were carried out to validate the proposed methods.