Real-time non-radiation-based navigation using 3D ultrasound for pedicle screw placement

BACKGROUND CONTEXT Spinal fusion surgeries increased by 137% between the years 1998–2008 and the requirement will continue to grow as the implications for fusion expand. Furthermore, total hospital charges, excluding charges for readmission, for 3.6 million spinal fusion surgeries, performed between 2001-2010 in United States, were more than $287 billion. Many such cases involved the use of pedicle screw fixation. Screw placement accuracy impacts fusion rate, adjacent level disease, and risk for neurovascular complications. Reported rates for screw misplacement, which required reoperation, were ranging up to 42%. Image guided surgical systems have provided means to improve screw placement accuracy, however, the clinical accuracy requirements for certain levels of spine still exceed the accuracy of current image guided surgical systems. Furthermore, most of these systems are based on the use of intra-operative two-dimensional (2D)/three-dimensional (3D) fluoroscopy imaging which operate with ionizing radiation. Finally, most of the available commercial systems are expensive resulting in a significant capital expense to the hospital, have a steep learning curve, and require new workflow to be established. PURPOSE The purpose of this work is to demonstrate that with improved modeling, segmentation and registration tools real-time 3D ultrasound can be used for spinal imaging and has potential to be used in spinal fusion surgeries. Our long-term goal is the successful integration of 3D ultrasound (US) as a standard of care intra-operative imaging modality for spinal fusion surgeries which has not been achieved so far. STUDY DESIGN/SETTING The computational methods developed during the project period were evaluated on in vivo ultrasound scans collected from healthy individuals. Data collection involved the use of a clinical grade ultrasound machine as well as a point of care wireless ultrasound device. Evaluations were performed against gold standard expert annotations. PATIENT SAMPLE A total of 27 healthy subjects were enrolled during the study. We have also collected retrospective spine CT scans in order to built a statistical spine shape model. OUTCOME MEASURES Evaluation studies were performed by measuring surface localization accuracy, classification accuracy and target registration accuracy. METHODS We have designed new artificial intelligence methods for segmentation, enhancement and classification of spine surfaces from ultrasound data. In order to provide a solution for manual operation of the ultrasound transducer during data collection, we have developed methods, based on deep learning, for automatic guidance of the ultrasound transducer to the correct scan plane. Finally, we have developed a novel machine learning-based point cloud registration method which does not require initialization to start the registration. RESULTS Our proposed methods outperform current state of the art in accuracy, robustness and processing time. For anatomical landmark localization we have obtained 0.1 mm (Standard deviation 0.2 mm) accuracy with a processing time of 54 milliseconds. We achieve submillimeter anatomical landmark localization accuracy. Sensitivity and specificity values for scan plane and vertebrae level classification were greater than 86%. CONCLUSIONS Our specific contribution in this work include: 1) a robust, accurate and fast spine bone enhancement and segmentation method from ultrasound data; 2) a multi-modal intra-operative image fusion strategy which can be applied to different orthopedic procedures as well. Future work will involve validation of our proposed methods on cadaver studies for pedicle screw guidance. The proposed ultrasound-based navigation system can also be used in conjunction with any existing 2D fluoroscopy/CT-nav-based navigation in order to provide additional augmentation and improve the overall accuracy of these systems while reducing the radiation exposure. FDA DEVICE/DRUG STATUS This abstract does not discuss or include any applicable devices or drugs.