Benjamin Wagner

Tissue thickness estimation for high precision head-tracking using a galvanometric laser scanner - a case study.

Marker-less optical head-tracking constitutes a comfortable alternative with no exposure to radiation for realtime monitoring in radiation therapy. Supporting information such as tissue thickness has the potential to improve spatial tracking accuracy. Here we study how accurate tissue thickness can be estimated from the near-infrared (NIR) backscatter obtained from laser scans. In a case study, optical data was recorded with a galvanometric laser scanner from three subjects. A tissue ground truth from MRI was robustly matched via customized bite blocks. We show that Gaussian Processes accurately model the relationship between NIR features and tissue thickness. They were able to predict the tissue thickness with less than 0.5 mm root mean square error. Individual scaling factors for all features and an additional incident angle feature had positive effects on this performance.

Preliminary study on optical feature detection for head tracking in radiation therapy

Marker-less tracking provides a non-invasive as well as comfortable approach to compensate for head motion in high precision radiotherapy. However, it suffers from a lack of point-to-point correspondences, typically requiring characteristic spatial landmarks to match point clouds. In this study, we show that cutaneous and subcutaneous structures can be uncovered using an 850 nm laser setup. For three subjects, we compare features extracted from camera images with MR scans serving as an anatomical ground truth. The results confirm the validity of the optically detected structures. The negative correlation between skin thickness and reflected light energy is likewise predicted by Monte-Carlo simulations and can be used to improve spatial point cloud matching. Tissue thickness and its facial structure can be predicted with submillimeter accuracy using a Support Vector regression machine. In addition, the optical measurements reveal the location of vessels that are not immediately visible in the MR scan. These promising findings highly encourage its application for a marker-less tracking system.

Data-Driven Learning for Calibrating Galvanometric Laser Scanners

State-of-the-art calibration very often constructs models motivated by a real-world device. Recently, artificial neural networks (ANNs) have been proposed as a more universal, accurate, and practical black-box approach. For a galvanometric triangulation device based on two mirrors, we embrace this proposal and set it into context with other supervised data-driven approaches: 1) ridge regression; 2) support vector regression; and 3) Gaussian processes. We show that they outperform available model-based approaches and yield similar performance compared with a memorizing lookup table calibration. The results demonstrate that an off-the-shelf usage of ANNs may run into generalization problems. Restricting the space of functions using kernel-based learning has proven to be advantageous. Finally, all approaches and distinct properties are discussed in a broader context, since each application entails differently relevant requirements for its calibration. This also holds for any calibration other than the considered triangulation device.

Tissue segmentation from head MRI: a ground truth validation for feature-enhanced tracking

Abstract Accuracy is essential for optical head-tracking in cranial radiotherapy. Recently, the exploitation of local patterns of tissue information was proposed to achieve a more robust registration. Here, we validate a ground truth for this information obtained from high resolution MRI scans. In five subjects we compared the segmentation accuracy of a semi-automatic algorithm with five human experts. While the algorithm segments the skin and bone surface with an average accuracy of less than 0.1 mm and 0.2 mm, respectively, the mean error on the tissue thickness was 0.17 mm. We conclude that this accuracy is a reasonable basis for extracting reliable cutaneous structures to support surface registration.

An Approach to Improve Accuracy of Optical Tracking Systems in Cranial Radiation Therapy.

This work presents a new method for the accurate estimation of soft tissue thickness based on near infrared (NIR) laser measurements. By using this estimation, our goal is to develop an improved non-invasive marker-less optical tracking system for cranial radiation therapy. Results are presented for three subjects and reveal an RMS error of less than 0.34 mm.

Calibration of Galvanometric Laser Scanners Using Statistical Learning Methods

Galvanometric laser scanners can be used for optical tracking. Model-based calibration of these systems is inaccurate and not adaptable to variations in the system. Therefore, a calibration method based on statistical learning methods is presented which directly incorporates the triangulation problem. We investigate linear regression as well as Artificial Neural Networks. The results are validated using (1) the cross-validated prediction accuracy within the calibration space, and (2) plane reconstruction accuracy. All statistical learning methods outperformed the model-based approach leading to an improvement of up to 74% for the cross-validated 3D root-mean-square error and 70-74% for the plane reconstruction. While the neural network achieved mean errors below 0.5 mm, the linear regression results suggest a good compromise between accuracy and computational load.

Ray interpolation for generic triangulation based on a galvanometric laser scanning system

Our research group is currently developing a new optical head tracking system for intracranial radiosurgery. The system uses infrared laser light to measure features of the soft tissue on the patients forehead which correlate with the thickness of the soft tissue. For future registration purposes, the system also acquires accurate pointwise reconstructions of the corresponding surface of the forehead by using triangulation. This paper proposes a method for the interpolation of laser rays based on a galvanometric laser scanning system. The advantage of the ray interpolation is that an alternative time consuming recalibration of the laser rays can be avoided. Experiments revealed a triangulation accuracy of around 0.1192 mm using ray interpolation.

Efficient estimation of tissue thicknesses using sparse approximation for Gaussian processes

Highly accurate localization of the human skull is vital in cranial radiotherapy. Marker-less optical head tracking provides a fast and accurate way to monitor this motion. Recent research has given evidence that marker-less tracking of the forehead benefits from tissue thickness information in addition to the 3D surface geometry. Using Gaussian Processes (GPs) tissue thickness is determined from optical backscatter of a sweeping laser. However, the computational complexity of the GPs scales cubically with the number of training samples. A full head scan contains 1024 points, whereas scans from several perspectives may be required for a comprehensive model for each subject. In five subjects, we thus evaluate sparse approximation methods to reduce the computational effort. We found a better - computation time versus root mean square error (RMSE) - tradeoff for a simple subset of data (SoD) technique. The increase of RMSE when dropping data was not found steep enough to justify the computational overhead of a better approximation by inducing point methods (namely FITC). Promising results were, however, obtained when clustering the training data before selecting the subset.

Analysis of feature stability for laser-based determination of tissue thickness

Localisation of the cranium is necessary for accurate stereotactic radiotherapy of malign lesions in the brain. This is achieved by immobilizing the patients head (typically by using thermoplastic masks, bite blocks or combinations thereof) and x-ray imaging to determine the actual position of the patient with respect to the treatment device. In previous work we have developed a novel method for marker-less and non-invasive tracking of the skull using a combination of laser-based surface triangulation and the analysis of backscattered feature patterns of a tightly collimated NIR laser beam scanned over the patients forehead. An HDR camera is coupled into the beam path of the laser scanning system to acquire one image per projected laser point. We have demonstrated that this setup is capable of accurately determining the tissue thickness for each triangulation point and consequently allows detecting the surface of the cranial bone with sub-millimetre accuracy. Typical clinical settings (treatment times of 15-90 min) require feature stability over time, since the determination of tissue thickness is achieved by machine learning methods trained on initial feature scans. We have collected initial scans of the forehead as well as long-term backscatter data (20 images per seconds over 30 min) from five subjects and extracted the relevant tissue features from the image streams. Based on the knowledge of the relationship between the tissue feature values and the tissue thickness, the analysis of the long-term data showed that the noise level is low enough to allow robust discrimination of tissue thicknesses of 0.5 mm.

Feasibility test of line sensors for optical tissue thickness estimation

Purpose Line sensors are cheap, fast and have high quantum effciencies. Here, we investigate whether these sensors can replace an area image sensor for the purpose of tissue thickness measurements. Material and Methods As part of a subject study high dynamic range (HDR) images of three subjects were acquired with an area image sensor. To simulate a line sensor as realistic as possible single or multiple lines were extracted from these HDR images. Thereby, horizontally extracted lines correspond to a parallel orientation of the line sensor relative to the incident angle of a laser beam. Vertically extracted lines correspond to an orthogonal orientation. Then, optical features were determined and converted into a tissue thickness using a machine learning algorithm. Results For the tested subjects the worst root mean square error (RMSE) of the learning process was 0:385 mm. The best RMSE was 0:222 mm. For all subjects, the mean RMSE and the standard deviation of RMSE values decreases with a larger number of extracted lines. The orientation of the line sensor turned out to be important for the RMSE. Vertically oriented line sensors achieve lower RMSEs than horizontally oriented sensors because of the influence of the incident angle. Furthermore, the head-pose of the subject seems to be important for the accuracy. Conclusion Line sensors deliver comparable results to previously analysed area image sensors. Nevertheless, the scattering of the values is higher and the size and orientation of the sensor and the head-pose have an influence on the RMSE of the learning process. Therefore, line sensors are feasible for tissue thickness estimation but they are a trade-off between accuracy and speed.