Catálogo de publicaciones - libros

Emphysema is characterized by the destruction and over distension of lung tissue, which manifest on high resolution computer tomography (CT) images as regions of low attenuation. Typically, it is diagnosed by clinical symptoms, physical examination, pulmonary function tests, and X-ray and CT imaging. In this paper we discuss a quantitative imaging approach to analyze emphysema which employs low-level segmentations of CT images that partition the data into perceptually relevant regions. We constructed multi-dimensional histograms of feature values computed over the image segmentation. For each region in the segmentation, we derive a rich set of feature measurements. While we can use any combination of physical and geometric features, we found that limiting the scope to two features – the mean attenuation across a region and the region area – is effective. The subject histogram is compared to a set of canonical histograms representative of various stages of emphysema using the Earth Mover’s Distance metric. Disease severity is assigned based on which canonical histogram is most similar to the subject histogram. Experimental results with 81 cases of emphysema at different stages of disease progression show good agreement against the reading of an expert radiologist.

In this paper, we propose a new dynamic learning framework that requires a small amount of labeled data in the beginning, then incrementally discovers informative unlabeled data to be hand-labeled and incorporates them into the training set to improve learning performance. This approach has great potential to reduce the training expense in many medical image analysis applications. The main contributions lie in a new strategy to combine confidence-rated classifiers learned on different feature sets and a robust way to evaluate the “informativeness” of each unlabeled example. Our framework is applied to the problem of classifying microscopic cell images. The experimental results show that 1) our strategy is more effective than simply multiplying the predicted probabilities, 2) the error rate of high-confidence predictions is much lower than the average error rate, and 3) hand-labeling informative examples with low-confidence predictions improves performance efficiently and the performance difference from hand-labeling all unlabeled data is very small.

Fibered confocal microscopy allows and imaging with cellular resolution. The potentiality of this imaging modality is extended in this work by using video mosaicing techniques. Two novelties are introduced. A robust estimator based on statistics for Riemannian manifolds is developed to find a globally consistent mapping of the input frames to a common coordinate system. A mosaicing framework using an efficient scattered data fitting method is proposed in order to take into account the non-rigid deformations and the irregular sampling implied by fibered confocal microscopy. Results on 50 images of a live mouse colon demonstrate the effectiveness of the proposed method.

The interaction of the microtubules with the cell cortex plays numerous critical roles in a cell. For instance, it directs vesicle delivery, and modulates membrane adhesions pivotal for cell movement as well as mitosis. Abnormal function of the microtubules is involved in cancer. An effective method to observe microtubule function adjacent to the cortex is TIRFM. To date most analysis of TIRFM images has been done by visual inspection and manual tracing. In this work we have developed a method to automatically process TIRFM images of microtubules so as to enable high throughput quantitative studies. The microtubules are extracted in terms of consecutive segments. The segments are described via Hamilton-Jacobi equations. Subsequently, the algorithm performs a limited reconstruction of the microtubules in 3D. Last, we evaluate our method with phantom as well as real TIRFM images of living cells.

A passive forward kinematics knee model was used to predict knee motion of a total joint replacement. Given a joint angle, maps of articular surfaces, and patient-specific ligament properties, this model predicted femorotibial contact locations based on the principle of ligament-strain minimization. The model was validated by physical experiments on a commonly implanted knee prosthesis, showing excellent correspondence between the model and actual physical motion. Results suggest that the knee prosthesis studied required an intact posterior cruciate ligament to induce the desirable roll-back motion, and that a single-bundle model of major knee ligaments generated kinematics similar to that of a multi-bundle ligament model. Implications are that a passive model may predict knee kinematics of a given patient, so it may be possible to optimize the implantation of a prosthesis intraoperatively.

Quantification of left ventricular (LV) deformation from 3D image sequences (4D data) is important for the assessment of myocardial viability, which can have important clinical implications. To date, feature information from either Magnetic Resonance, computed tomographic or echocardiographic image data has been assembled with the help of different interpolative models to estimate LV deformation. These models typically are designed to be computationally efficient (e.g. regularizing strategies using B-splines) or more physically realistic (e.g. finite element approximations to biomechanical models), but rarely incorporate both notions. In this paper, we combine an approach to the extraction and matching of image-derived point features based on local shape properties with a boundary element model. This overall scheme is intended to be both computationally efficient and physically realistic. In order to illustrate this, we compute strains using our method on canine 4D MR image sequences and compare the results to those found from a B-spline-based method (termed extended free-form deformation (EFFD)) and a method based on finite elements (FEM). All results are compared to displacements found using implanted markers, taken to be a gold standard.

A finite-volume model of the cerebrospinal fluid (CSF) system encompassing the third ventricle and the aqueduct of Sylvius was used to reconstruct CSF velocity and pressure fields based on MRI data. The flow domain geometry was obtained through segmentation of MRI brain anatomy scans. The movement of the domain walls was interpolated from brain motion MRI scans. A constant pressure boundary condition (BC) was specified at the foramina of Monro. A transient velocity BC reconstructed from velocimetric MRI scans was employed at the inferior end of the aqueduct of Sylvius. It could be shown that a combination of MRI scans and computational fluid dynamics (CFD) simulation can be used to reconstruct the flow field in the third ventricle. Pre-interventional knowledge of patient-specific CSF flow has the potential to improve neurosurgical interventions such as shunt placement in case of hydrocephalus.

Tissue engineering is a discipline at the leading edge of the field of computer assisted intervention. This multidisciplinary engineering science is based on the notion of design and fabrication of scaffolds- porous, three-dimensional biomimetic structures that, on implantation, provide a viable environment to recuperate and regenerate damaged cells. Existing CAD-based approaches produce porous labyrinths with straight edges. The biomorphic geometry that mimics the substrate would be one that is continuous through all space, partitioned into two not-necessarily-equal sub-spaces by a non-intersecting, two-sided surface. are ideal to describe such a space. We present results on the premier attempt in computer controlled fabrication and mechanical characterization of Triply Periodic Minimal Surfaces [TPMS]. This initiative is a significant step to link Schwann’s 1838 cell theory with Schwarz’s discovery of TPMS in 1865 to fabricate the previously elusive optimal biomorphic tissue analogs.

Robotic surgical systems such as Intuitive Surgical’s da Vinci system provide a rich source of motion and video data from surgical procedures. In principle, this data can be used to evaluate surgical skill, provide surgical training feedback, or document essential aspects of a procedure. If processed online, the data can be used to provide context-specific information or motion enhancements to the surgeon. However, in every case, the key step is to relate recorded motion data to a model of the procedure being performed. This paper examines our progress at developing techniques for “parsing” raw motion data from a surgical task into a labelled sequence of surgical gestures. Our current techniques have achieved > 90% fully automated recognition rates on 15 datasets.

We present daVinci Canvas: a telerobotic surgical system with integrated robot-assisted laparoscopic ultrasound capability. DaVinci Canvas consists of the integration of a rigid laparoscopic ultrasound probe with the daVinci robot, video tracking of ultrasound probe motions, endoscope and ultrasound calibration and registration, autonomous robot motions, and the display of registered 2D and 3D ultrasound images. Although we used laparoscopic liver cancer surgery as a focusing application, our broader aim was the development of a versatile system that would be useful for many procedures.