Ferdinand van der Heijden

Efficient adaptive density estimation per image pixel for the task of background subtraction

We analyze the computer vision task of pixel-level background subtraction. We present recursive equations that are used to constantly update the parameters of a Gaussian mixture model and to simultaneously select the appropriate number of components for each pixel. We also present a simple non-parametric adaptive density estimation method. The two methods are compared with each other and with some previously proposed algorithms.

Evaluation of prostate segmentation algorithms for MRI: the PROMISE12 challenge

Prostate MRI image segmentation has been an area of intense research due to the increased use of MRI as a modality for the clinical workup of prostate cancer. Segmentation is useful for various tasks, e.g. to accurately localize prostate boundaries for radiotherapy or to initialize multi-modal registration algorithms. In the past, it has been difficult for research groups to evaluate prostate segmentation algorithms on multi-center, multi-vendor and multi-protocol data. Especially because we are dealing with MR images, image appearance, resolution and the presence of artifacts are affected by differences in scanners and/or protocols, which in turn can have a large influence on algorithm accuracy. The Prostate MR Image Segmentation (PROMISE12) challenge was setup to allow a fair and meaningful comparison of segmentation methods on the basis of performance and robustness. In this work we will discuss the initial results of the online PROMISE12 challenge, and the results obtained in the live challenge workshop hosted by the MICCAI2012 conference. In the challenge, 100 prostate MR cases from 4 different centers were included, with differences in scanner manufacturer, field strength and protocol. A total of 11 teams from academic research groups and industry participated. Algorithms showed a wide variety in methods and implementation, including active appearance models, atlas registration and level sets. Evaluation was performed using boundary and volume based metrics which were combined into a single score relating the metrics to human expert performance. The winners of the challenge where the algorithms by teams Imorphics and ScrAutoProstate, with scores of 85.72 and 84.29 overall. Both algorithms where significantly better than all other algorithms in the challenge (p<0.05) and had an efficient implementation with a run time of 8min and 3s per case respectively. Overall, active appearance model based approaches seemed to outperform other approaches like multi-atlas registration, both on accuracy and computation time. Although average algorithm performance was good to excellent and the Imorphics algorithm outperformed the second observer on average, we showed that algorithm combination might lead to further improvement, indicating that optimal performance for prostate segmentation is not yet obtained. All results are available online at http://promise12.grand-challenge.org/.

Improving the selection of feature points for tracking

The problem considered in this paper is how to select the feature points (in practice, small image patches are used) in an image from an image sequence, such that they can be tracked adequately further through the sequence. Usually, the tracking is performed by some sort of local search method looking for a similar patch in the next image in the sequence. Therefore, it would be useful if we could estimate “the size of the convergence region” for each image patch. There is a smaller chance of error when calculating the displacement for an image patch with a large convergence region than for an image patch with a small convergence region. Consequently, the size of the convergence region can be used as a proper goodness measure for a feature point. For the standard Kanade-Lucas-Tomasi (KLT) tracking method, we propose a simple and fast way to approximate the convergence region for an image patch. In the experimental part, we test our hypothesis on a large set of real data.

Object recognition with stereo vision and geometric hashing

In this paper we demonstrate a method to recognize 3D objects and to estimate their pose. For that purpose we use a combination of stereo vision and geometric hashing. Stereo vision is used to generate a large number of 3D low level features, of which many are spurious because at that stage of the process the correspondence problem is not solved as yet. However, geometric hashing is used to discriminate the true features from the spurious one. Geometric hashing is also the basis of a voting mechanism for the recognition of the objects in the scene. The speed of the geometric hashing algorithm helps to overcome the computational burden imposed by the correspondence problem in stereo vision. We look at different hash strategies using both points and lines features and compare our 3D approach to a recognition system based on 2D features. Experiments show that, although our 3D approach generates much more spurious scene features, it is just as fast and more reliable than the 2D system.

Time-of-flight estimation based on covariance models

We address the problem of estimating the time-of-flight (ToF) of a waveform that is disturbed heavily by additional reflections from nearby objects. These additional reflections cause interference patterns that are difficult to predict. The introduction of a model for the reflection in terms of a non-stationary auto-covariance function leads to a new estimator for the ToF of an acoustic tone burst. This estimator is a generalization of the well known matched filter. In many practical circumstances, for instance beacon-based position estimation in indoor situations, lack of knowledge of the additional reflections can lead to large estimation errors. Experiments show that the application of the new estimator can reduce these errors by a factor of about four. The cost of this improvement is an increase in computational complexity by a factor of about seven.

Music playlist generation by assimilating GMMs into SOMs

A method for music playlist generation, using assimilated Gaussian mixture models (GMMs) in self organizing maps (SOMs) is presented. Traditionally, the neurons in a SOM are represented by vectors, but in this paper we propose to use GMMs instead. To this end, we introduce a method to adapt a GMM such that its distance to a second GMM decreases at a controllable rate. Self organization is demonstrated using a small music database and a music classification task.

Predicting 3D lip shapes using facial surface EMG

Aim The aim of this study is to prove that facial surface electromyography (sEMG) conveys sufficient information to predict 3D lip shapes. High sEMG predictive accuracy implies we could train a neural control model for activation of biomechanical models by simultaneously recording sEMG signals and their associated motions. Materials and methods With a stereo camera set-up, we recorded 3D lip shapes and simultaneously performed sEMG measurements of the facial muscles, applying principal component analysis (PCA) and a modified general regression neural network (GRNN) to link the sEMG measurements to 3D lip shapes. To test reproducibility, we conducted our experiment on five volunteers, evaluating several sEMG features and window lengths in unipolar and bipolar configurations in search of the optimal settings for facial sEMG. Conclusions The errors of the two methods were comparable. We managed to predict 3D lip shapes with a mean accuracy of 2.76 mm when using the PCA method and 2.78 mm when using modified GRNN. Whereas performance improved with shorter window lengths, feature type and configuration had little influence.

A New Accurate 3D Measurement Tool to Assess the Range of Motion of the Tongue in Oral Cancer Patients: A Standardized Model

In oral cancer treatment, function loss such as speech and swallowing deterioration can be severe, mostly due to reduced lingual mobility. Until now, there is no standardized measurement tool for tongue mobility and pre-operative prediction of function loss is based on expert opinion instead of evidence based insight. The purpose of this study was to assess the reliability of a triple-camera setup for the measurement of tongue range of motion (ROM) in healthy adults and its feasibility in patients with partial glossectomy. A triple-camera setup was used, and 3D coordinates of the tongue in five standardized tongue positions were achieved in 15 healthy volunteers. Maximum distances between the tip of the tongue and the maxillary midline were calculated. Each participant was recorded twice, and each movie was analysed three times by two separate raters. Intrarater, interrater and test–retest reliability were the main outcome measures. Secondly, feasibility of the method was tested in ten patients treated for oral tongue carcinoma. Intrarater, interrater and test–retest reliability all showed high correlation coefficients of >0.9 in both study groups. All healthy subjects showed perfect symmetrical tongue ROM. In patients, significant differences in lateral tongue movements were found, due to restricted tongue mobility after surgery. This triple-camera setup is a reliable measurement tool to assess three-dimensional information of tongue ROM. It constitutes an accurate tool for objective grading of reduced tongue mobility after partial glossectomy.

Predicting 3D lip movement using facial sEMG: a first step towards estimating functional and aesthetic outcome of oral cancer surgery

In oral cancer, loss of function due to surgery can be unacceptable, designating the tumour as functionally inoperable. Other curative treatments can then be considered. Currently, predictions of these functional consequences are subjective and unreliable. We want to create patient-specific models to improve and objectify these predictions. A first step was taken by controlling a 3D lip model with volunteer-specific sEMG activities. We focus on the lips first, because they are essential for speech, oral food transport, and facial mimicry. Besides, they are more accessible to measurements than intraoral organs. 3D lip movement and corresponding sEMG activities are measured in five healthy volunteers, who performed 19 instructions repeatedly, to create a quantitative lip model by establishing the relationship between sEMG activities of eight facial muscles bilaterally on the input side and the corresponding 3D lip displacements on the output side. The relationship between 3D lip movement and sEMG activities was accommodated in a state-space model. A good relationship between sEMG activities and 3D lip movement was established with an average root mean square error of 2.43 mm for the first-order system and 2.46 mm for the second-order system. This information can be incorporated into biomechanical models to further personalise functional outcome assessment after treatment.

The MIRIAM Robot: A Novel Robotic System for MR-Guided Needle Insertion in the Prostate

Early prostate cancer detection and treatment are of major importance to reduce mortality rate. magnetic resonance (MR) imaging provides images of the prostate where an early stage lesion can be visualized. The use of robotic systems for MR-guided interventions in the prostate allows us to improve the clinical outcomes of procedures such as biopsy and brachytherapy. This work presents a novel MR-conditional robot for prostate interventions. The minimally invasive robotics in an magnetic resonance imaging environment (MIRIAM) robot has 9 degrees-of-freedom (DoF) used to steer and fire a biopsy needle. The needle guide is positioned against the perineum by a 5 DoF parallel robot driven by piezoelectric motors. A 4 DoF needle driver inserts, rotates and fires the needle during the procedure. Piezoelectric motors are used to insert and rotate the needle, while pneumatic actuation is used to fire the needle. The MR-conditional design of the robot and the needle insertion controller are presented. MR compatibility tests using T2 imaging protocol are performed showing a SNR reduction of 25% when the robot is operational within the MR scanner. Experiments inserting a biopsy needle toward a physical target resulted in an average targeting error of 1.84mm. Our study presents a novel MR-conditional robot and demonstrated the ability to perform MR-guided needle-based interventions in soft-tissue phantoms. Moreover, the image distortion analysis indicates that no visible image deterioration is induced by the robot.