Convolutional Neural Networks (CNN) are state-of-the-art models for many image classification tasks. However, to recognize cancer subtypes automatically, training a CNN on gigapixel resolution Whole Slide Tissue Images (WSI) is currently computationally impossible. The differentiation of cancer subtypes is based on cellular-level visual features observed on image patch scale. Therefore, we argue that in this situation, training a patch-level classifier on image patches will perform better than or similar to an image-level classifier. The challenge becomes how to intelligently combine patch-level classification results and model the fact that not all patches will be discriminative. We propose to train a decision fusion model to aggregate patch-level predictions given by patch-level CNNs, which to the best of our knowledge has not been shown before. Furthermore, we formulate a novel Expectation-Maximization (EM) based method that automatically locates discriminative patches robustly by utilizing the spatial relationships of patches. We apply our method to the classification of glioma and non-small-cell lung carcinoma cases into subtypes. The classification accuracy of our method is similar to the inter-observer agreement between pathologists. Although it is impossible to train CNNs on WSIs, we experimentally demonstrate using a comparable non-cancer dataset of smaller images that a patch-based CNN can outperform an image-based CNN.

No abstract

We address the problem of building watertight 3D models from surfaces that contain holes-for example, sets of range scans that observe most but not all of a surface. We specifically address situations in which the holes are too geometrically and topologically complex to fill using triangulation algorithms. Our solution begins by constructing a signed distance function, the zero set of which defines the surface. Initially, this function is defined only in the vicinity of observed surfaces. We then apply a diffusion process to extend this function through the volume until its zero set bridges whatever holes may be present. If additional information is available, such as known-empty regions of space inferred from the lines of sight to a 3D scanner, it can be incorporated into the diffusion process. Our algorithm is simple to implement, is guaranteed to produce manifold non-interpenetrating surfaces, and is efficient to run on large datasets because computation is limited to areas near holes. By showing results for complex range scans, we demonstrate that our algorithm produces hole-free surfaces that are plausible, visually acceptable, and usually close to the intended geometry.

a) (b) (c) (d) Figure 1: Rendering comparisons. (a)Rendering of 3D scanned range image, (b) same scanned geometry, augmented with a measured normal-map (from photometric stereo), (c) our hybrid surface reconstruction, which combines both position and normal constraints, (d) photograph. Renderings in this paper do not use color information in order to focus on geometric aspects. Note how our method eliminates noise from the range image while introducing real detail. The surface normals are of the same quality or better than those from photometric stereo, while most of the low-frequency bias has been eliminated. AbstractRange scanning, manual 3D editing, and other modeling approaches can provide information about the geometry of surfaces in the form of either 3D positions (e.g., triangle meshes or range images) or orientations (normal maps or bump maps). We present an algorithm that combines these two kinds of estimates to produce a new surface that approximates both. Our formulation is linear, allowing it to operate efficiently on complex meshes commonly used in graphics. It also treats high-and low-frequency components separately, allowing it to optimally combine outputs from data sources such as stereo triangulation and photometric stereo, which have different error-vs.-frequency characteristics. We demonstrate the ability of our technique to both recover high-frequency details and avoid low-frequency bias, producing surfaces that are more widely applicable than position or orientation data alone.

Much of the research on video-based human motion capture assumes the body shape is known