Deep convolutional neural networks for regular texture recognition.

Regular textures are frequently found in man-made environments and some biological and physical images. There are a wide range of applications for recognizing and locating regular textures. In this work, we used deep convolutional neural networks (CNNs) as a general method for modelling and classifying regular and irregular textures. We created a new regular texture database and investigated two sets of deep CNNs-based methods for regular and irregular texture classification. First, the classic CNN models (e.g. inception, residual network, etc.) were used in a standard way. These two-class CNN classifiers were trained by fine-tuning networks using our new regular texture database. Next, we transformed the trained filter features of the last convolutional layer into a vector representation using Fisher Vector pooling (FV). Such representations can be efficiently used for a wide range of machine learning tasks such as classification or clustering, thus more transferable from one domain to another. Our experiments show that the standard CNNs attained sufficient accuracy for regular texture recognition tasks. The Fisher representations combined with support vector machine (SVM) also showed high performance for regular and irregular texture classification. We also find CNNs performs sub-optimally for long-range patterns, despite the fact that their fully-connected layers pool local features into a global image representation.

Building a three-dimensional model of the upper gastrointestinal tract for computer simulations of swallowing.

We aimed to provide realistic three-dimensional (3D) models to be used in numerical simulations of peristaltic flow in patients exhibiting difficulty in swallowing, also known as dysphagia. To this end, a 3D model of the upper gastrointestinal tract was built from the color cryosection images of the Visible Human Project dataset. Regional color heterogeneities were corrected by centering local histograms of the image difference between slices. A voxel-based model was generated by stacking contours from the color images. A triangle mesh was built, smoothed and simplified. Visualization tools were developed for browsing the model at different stages and for virtual endoscopy navigation. As result, a computer model of the esophagus and the stomach was obtained, mainly for modeling swallowing disorders. A central-axis curve was also obtained for virtual navigation and to replicate conditions relevant to swallowing disorders modeling. We show renderings of the model and discuss its use for simulating swallowing as a function of bolus rheological properties. The information obtained from simulation studies with our model could be useful for physicians in selecting the correct nutritional emulsions for patients with dysphagia.

Diagnostic radiograph based 3D bone reconstruction framework: application to osteotomy surgical planning.

Pre-operative planning in orthopedic surgery is essential to identify the optimal surgical considerations for each patient-specific case. The planning for osteotomy is presently conducted through two-dimensional (2D) radiographs, where the surgeon has to mentally visualize the bone deformity. This is due to direct three-dimensional (3D) imaging modalities such as Computed Tomography (CT) still being restricted to a minority of complex orthopedic procedures. This paper presents a novel 3D bone reconstruct technique, through bi-planar 2D radiographic images. The reconstruction will be pertinent to osteotomy surgical diagnostics and planning. The framework utilizes a generic 3D model of the bone of interest to obtain the anatomical topology information. A 2D non-rigid registration is performed between the projected contours of this generic 3D model and extracted edges of the X-ray image to identify the planar customization required. Subsequently a free-form deformation based manipulation is conducted to customize the overall 3D bone shape.

Segmentation of radiographic images under topological constraints: application to the femur.

PURPOSE: A framework for radiographic image segmentation under topological control based on two-dimensional (2D) image analysis was developed. The system is intended for use in common radiological tasks including fracture treatment analysis, osteoarthritis diagnostics and osteotomy management planning. METHODS: The segmentation framework utilizes a generic three-dimensional (3D) model of the bone of interest to define the anatomical topology. Non-rigid registration is performed between the projected contours of the generic 3D model and extracted edges of the X-ray image to achieve the segmentation. For fractured bones, the segmentation requires an additional step where a region-based active contours curve evolution is performed with a level set Mumford-Shah method to obtain the fracture surface edge. The application of the segmentation framework to analysis of human femur radiographs was evaluated. The proposed system has two major innovations. First, definition of the topological constraints does not require a statistical learning process, so the method is generally applicable to a variety of bony anatomy segmentation problems. Second, the methodology is able to handle both intact and fractured bone segmentation. RESULTS: Testing on clinical X-ray images yielded an average root mean squared distance (between the automatically segmented femur contour and the manual segmented ground truth) of 1.10 mm with a standard deviation of 0.13 mm. The proposed point correspondence estimation algorithm was benchmarked against three state-of-the-art point matching algorithms, demonstrating successful non-rigid registration for the cases of interest. CONCLUSIONS: A topologically constrained automatic bone contour segmentation framework was developed and tested, providing robustness to noise, outliers, deformations and occlusions.

A mesh-free mechanical model of the upper gastrointestinal system.

Realistic behavior in Computer Simulation of biological system (e.g. humans organs) is essential to 3D modeling in medicine. In order to improve realistic responses of 3D organ model it is essential to use mechanical models that can deal with multiple objects internal and external interactions in a reasonable time frame. We will apply the Smooth Particles Hydrodynamics (SPH) to model the esophagus and the stomach, thus constructing a physical background for interaction. We used a multilayer model of particles related to a single triangle mesh. Each particle layers represent distinct biological tissues of the esophagus and the stomach.