Michael Tschannen

A learning-based approach for fast and robust vessel tracking in long ultrasound sequences.

We propose a learning-based method for robust tracking in long ultrasound sequences for image guidance applications. The framework is based on a scale-adaptive block-matching and temporal realignment driven by the image appearance learned from an initial training phase. The latter is introduced to avoid error accumulation over long sequences. The vessel tracking performance is assessed on long 2D ultrasound sequences of the liver of 9 volunteers under free breathing. We achieve a mean tracking accuracy of 0.96 mm. Without learning, the error increases significantly (2.19 mm, p<0.001).

Subspace clustering of dimensionality-reduced data

Subspace clustering refers to the problem of clustering unlabeled high-dimensional data points into a union of low-dimensional linear subspaces, assumed unknown. In practice one may have access to dimensionality-reduced observations of the data only, resulting, e.g., from “undersampling” due to complexity and speed constraints on the acquisition device. More pertinently, even if one has access to the high-dimensional data set it is often desirable to first project the data points into a lower-dimensional space and to perform the clustering task there; this reduces storage requirements and computational cost. The purpose of this paper is to quantify the impact of dimensionality-reduction through random projection on the performance of the sparse subspace clustering (SSC) and the thresholding based subspace clustering (TSC) algorithms. We find that for both algorithms dimensionality reduction down to the order of the subspace dimensions is possible without incurring significant performance degradation. The mathematical engine behind our theorems is a result quantifying how the affinities between subspaces change under random dimensionality reducing projections.

Deep structured features for semantic segmentation

We propose a highly structured neural network architecture for semantic segmentation with an extremely small model size, suitable for low-power embedded and mobile platforms. Specifically, our architecture combines i) a Haar wavelet-based tree-like convolutional neural network (CNN), ii) a random layer realizing a radial basis function kernel approximation, and iii) a linear classifier. While stages i) and ii) are completely pre-specified, only the linear classifier is learned from data. We apply the proposed architecture to outdoor scene and aerial image semantic segmentation and show that the accuracy of our architecture is competitive with conventional pixel classification CNNs. Furthermore, we demonstrate that the proposed architecture is data efficient in the sense of matching the accuracy of pixel classification CNNs when trained on a much smaller data set.

Nonparametric nearest neighbor random process clustering

We consider the problem of clustering noisy finite-length observations of stationary ergodic random processes according to their nonparametric generative models without prior knowledge of the model statistics and the number of generative models. Two algorithms, both using the L1-distance between estimated power spectral densities (PSDs) as a measure of dissimilarity, are analyzed. The first algorithm, termed nearest neighbor process clustering (NNPC), to the best of our knowledge, is new and relies on partitioning the nearest neighbor graph of the observations via spectral clustering. The second algorithm, simply referred to as k-means (KM), consists of a single k-means iteration with farthest point initialization and was considered before in the literature, albeit with a different measure of dissimilarity and with asymptotic performance results only. We show that both NNPC and KM succeed with high probability under noise and even when the generative process PSDs overlap significantly, all provided that the observation length is sufficiently large. Our results quantify the tradeoff between the overlap of the generative process PSDs, the noise variance, and the observation length. Finally, we present numerical performance results for synthetic and real data.

Regression forest-based automatic estimation of the articular margin plane for shoulder prosthesis planning

In shoulder arthroplasty, the proximal humeral head is resected by sawing along the cartilage-bone transition and replaced by a prosthetic implant. The resection plane, called articular margin plane (AMP), defines the orientation, position and size of the prosthetic humeral head in relation to the humeral shaft. Therefore, the correct definition of the AMP is crucial for the computer-assisted preoperative planning of shoulder arthroplasty. We present a fully automated method for estimating the AMP relying only on computed tomography (CT) images of the upper arm. It consists of two consecutive steps, each of which uses random regression forests (RFs) to establish a direct mapping from the CT image to the AMP parameters. In the first step, image intensities serve as features to compute a coarse estimate of the AMP. The second step builds upon this estimate, calculating a refined AMP using novel feature types that combine a bone enhancing sheetness measure with ray features. The proposed method was evaluated on a dataset consisting of 72 CT images of upper arm cadavers. A mean localization error of 2.40mm and a mean angular error of 6.51° was measured compared to manually annotated ground truth.