In modern health care, many diagnostic and therapeutic procedures are performed using flexible endoscopes. One of the main applications is flexible bronchoscopy, and in particular transbronchial needle aspiration is routinely performed e.g. for diagnosis of pulmonary cancer. While expert bronchoscopists are well able to orient themselves, there will be significant benefit for operators, patients, and workflow, when information from preoperative imaging is translated to the operating room in an intelligent manner. For transbronchial needle aspiration, as mentioned above, the location of the biopsy target (solitary pulmonary nodules or lymph nodes) is precisely known from CT. However, if it is not visible from inside the airways or not reachable using video guidance, then navigation systems can provide invaluable support. We use an electromagnetic tracking system and tracking sensors embedded at the tip of the bronchoscope, in order to track the bronchoscope within the patient. After initial point-based registration, electromagnetic tracking measurements can be used as approximations of the scope’s position and orientation at each point in time of the surgery. However, due to hardware imperfections, respiratory motion, and distortions of the electromagnetic field, tracking errors are introduced, which need to be detected and corrected. Since pre-operative CT data is available, we may create "virtual bronchoscopy" images at the scope’s known approximate pose (i.e. position and orientation), and then we may optimize this pose for best possible similarity with the real bronchoscopy image. Such methods have already been proposed, but one shortcoming is that small changes in the electromagnetic tracking measurements as input to the optimization may result in large changes of the corrected output. Such amplified jitter leads e.g. to a shaky visualisation in the case of augmented reality overlays. Thus, we propose modelling the corrected output as a continuous curve, since the physical motion of the bronchoscope will also be continuous. We show that within such a framework, we can easily include other prior knowledge like the mentioned image similarity, but also the amount of deviation from the original electromagnetic tracking measurements or regularisations of the curve shape. The proposed method is evaluated using a dynamic phantom data set and an ex-vivo porcine lung explants with cyclic motion. Results are compared to ground truth obtained from multiple human experts. Since ground truth is provided multiple times by multiple experts, we can also assess the variance within each expert (between 1.4 and 1.7 mm for the phantom data set, 2.3 mm for the porcine lung) and among the experts (1.3 mm), which may also serve as a baseline for the performance of human experts. In comparison to the ground truth data, the accuracy of the proposed method is equivalent to previous approaches, as is computational efficiency. However, smoothness in terms of intra-frame distance, i.e. change between successive frames, improved from 3.2-3.7 mm to 1.2 mm for the proposed method. This quantitative result is confirmed by visualization (cf. Figure 1), which shows a smooth, continuous, and physically plausible trajectory. We conclude that the proposed method provides a smoother motion than the state-of-art, while maintaining equivalent accuracy and computational efficiency. This result was verified using a thorough evaluation with ground truth data provided by human experts.     «

In modern health care, many diagnostic and therapeutic procedures are performed using flexible endoscopes. One of the main applications is flexible bronchoscopy, and in particular transbronchial needle aspiration is routinely performed e.g. for diagnosis of pulmonary cancer. While expert bronchoscopists are well able to orient themselves, there will be significant benefit for operators, patients, and workflow, when information from preoperative imaging is translated to the operating room in an...     »