Rahul Bhotika

Flexible flow for 3D nonrigid tracking and shape recovery

We introduce linear methods for model-based tracking of nonrigid 3D objects and for acquiring such models from video. 3D motions and flexions are calculated directly from image intensities without information-lossy intermediate results. Measurement uncertainty is quantified and fully propagated through the inverse model to yield posterior mean (PM) and/or mode (MAP) pose estimates. A Bayesian framework manages uncertainty, accommodates priors, and gives confidence measures. We obtain highly accurate and robust closed-form estimators by minimizing information loss from non-reversible (inner-product and least-squares) operations, and, when unavoidable, performing such operations with the appropriate error norm. For model acquisition, we show how to refine a crude or generic model to fit the video subject. We demonstrate with tracking, model refinement, and super-resolution texture lifting from low-quality low-resolution video.

Joint Recognition of Complex Events and Track Matching

We present a novel method for jointly performing recognition of complex events and linking fragmented tracks into coherent, long-duration tracks. Many event recognition methods require highly accurate tracking, and may fail when tracks corresponding to event actors are fragmented or partially missing. However, these conditions occur frequently from occlusions, traffic and tracking errors. Recently, methods have been proposed for linking track fragments from multiple objects under these difficult conditions. Here, we develop a method for solving these two problems jointly. A hypothesized event model, represented as a Dynamic Bayes Net, supplies data-driven constraints on the likelihood of proposed track fragment matches. These event-guided constraints are combined with appearance and kinematic constraints used in the previous track linking formulation. The result is the most likely track linking solution given the event model, and the highest event score given all of the track fragments. The event model with the highest score is determined to have occurred, if the score exceeds a threshold. Results demonstrated on a busy scene of airplane servicing activities, where many non-event movers and long fragmented tracks are present, show the promise of the approach to solving the joint problem.

Multi-material decomposition of spectral CT images

Spectral Computed Tomography (Spectral CT), and in particular fast kVp switching dual-energy computed tomography, is an imaging modality that extends the capabilities of conventional computed tomography (CT). Spectral CT enables the estimation of the full linear attenuation curve of the imaged subject at each voxel in the CT volume, instead of a scalar image in Hounsfield units. Because the space of linear attenuation curves in the energy ranges of medical applications can be accurately described through a two-dimensional manifold, this decomposition procedure would be, in principle, limited to two materials. This paper describes an algorithm that overcomes this limitation, allowing for the estimation of N-tuples of material-decomposed images. The algorithm works by assuming that the mixing of substances and tissue types in the human body has the physicochemical properties of an ideal solution, which yields a model for the density of the imaged material mix. Under this model the mass attenuation curve of each voxel in the image can be estimated, immediately resulting in a material-decomposed image triplet. Decomposition into an arbitrary number of pre-selected materials can be achieved by automatically selecting adequate triplets from an application-specific material library. The decomposition is expressed in terms of the volume fractions of each constituent material in the mix; this provides for a straightforward, physically meaningful interpretation of the data. One important application of this technique is in the digital removal of contrast agent from a dual-energy exam, producing a virtual nonenhanced image, as well as in the quantification of the concentration of contrast observed in a targeted region, thus providing an accurate measure of tissue perfusion.

Model-Based analysis of local shape for lesion detection in CT scans

Thin-slice computer tomography provides high-resolution images that facilitate the diagnosis of early-stage lung cancer. However, the sheer size of the CT volumes introduces variability in radiological readings, driving the need for automated detection systems. The main contribution of this paper is a technique for combining geometric and intensity models with the analysis of local curvature for detecting pulmonary lesions in CT. The local shape at each voxel is represented via the principal curvatures of its associated isosurface without explicitly extracting the isosurface. The comparison of these curvatures to values derived from analytical shape models is then used to label the voxel as belonging to particular anatomical structures, e.g., nodules or vessels. The algorithm was evaluated on 242 CT exams with expert-determined ground truth. The performance of the algorithm is quantified by free-response receiver-operator characteristic curves, as well as by its potential for improvement in radiologist sensitivity.

Effective atomic number accuracy for kidney stone characterization using spectral CT

The clinical application of Gemstone Spectral ImagingTM, a fast kV switching dual energy acquisition, is explored in the context of noninvasive kidney stone characterization. Utilizing projection-based material decomposition, effective atomic number and monochromatic images are generated for kidney stone characterization. Analytical and experimental measurements are reported and contrasted. Phantoms were constructed using stone specimens extracted from patients. This allowed for imaging of the different stone types under similar conditions. The stone specimens comprised of Uric Acid, Cystine, Struvite and Calcium-based compositions. Collectively, these stone types span an effective atomic number range of approximately 7 to 14. While Uric Acid and Calcium based stones are generally distinguishable in conventional CT, stone compositions like Cystine and Struvite are difficult to distinguish resulting in treatment uncertainty. Experimental phantom measurements, made under increasingly complex imaging conditions, illustrate the impact of various factors on measurement accuracy. Preliminary clinical studies are reported.

Automated liver lesion characterization using fast kVp switching dual energy computed tomography imaging

Hypodense metastases are not always completely distinguishable from benign cysts in the liver using conventional Computed Tomography (CT) imaging, since the two lesion types present with overlapping intensity distributions due to similar composition as well as other factors including beam hardening and patient motion. This problem is extremely challenging for small lesions with diameter less than 1 cm. To accurately characterize such lesions, multiple follow-up CT scans or additional Positron Emission Tomography or Magnetic Resonance Imaging exam are often conducted, and in some cases a biopsy may be required after the initial CT finding. Gemstone Spectral Imaging (GSI) with fast kVp switching enables projection-based material decomposition, offering the opportunity to discriminate tissue types based on their energy-sensitive material attenuation and density. GSI can be used to obtain monochromatic images where beam hardening is reduced or eliminated and the images come inherently pre-registered due to the fast kVp switching acquisition. We present a supervised learning method for discriminating between cysts and hypodense liver metastases using these monochromatic images. Intensity-based statistical features extracted from voxels inside the lesion are used to train optimal linear and nonlinear classifiers. Our algorithm only requires a region of interest within the lesion in order to compute relevant features and perform classification, thus eliminating the need for an accurate segmentation of the lesion. We report classifier performance using M-fold cross-validation on a large lesion database with radiologist-provided lesion location and labels as the reference standard. Our results demonstrate that (a) classification using a single projection-based spectral CT image, i.e., a monochromatic image at a specified keV, outperforms classification using an image-based dual energy CT pair, i.e., low and high kVp images derived from the same fast kVp acquisition and (b) classification using monochromatic images can achieve very high accuracy in separating benign liver cysts and metastases, especially for small lesions.

Lung nodule detection via Bayesian voxel labeling

This paper describes a system for detecting pulmonary nodules in CT images. It aims to label individual image voxels in accordance to one of a number of anatomical (pulmonary vessels or junctions), pathological (nodules), or spurious (noise) events. The approach is orthodoxly Bayesian, with particular care taken in the objective establishment of prior probabilities and the incorporation of relevant medical knowledge. We provide, under explicit modeling assumptions, closed-form expressions for all the probability distributions involved. The technique is applied to real data, and we present a discussion of its performance.

Part-Based local shape models for colon polyp detection

This paper presents a model-based technique for lesion detection in colon CT scans that uses analytical shape models to map the local shape curvature at individual voxels to anatomical labels. Local intensity profiles and curvature information have been previously used for discriminating between simple geometric shapes such as spherical and cylindrical structures. This paper introduces novel analytical shape models for colon-specific anatomy, viz. folds and polyps, built by combining parts with simpler geometric shapes. The models better approximate the actual shapes of relevant anatomical structures while allowing the application of model-based analysis on the simpler model parts. All parameters are derived from the analytical models, resulting in a simple voxel labeling scheme for classifying individual voxels in a CT volume. The algorithms performance is evaluated against expert-determined ground truth on a database of 42 scans and performance is quantified by free-response receiver-operator curves.

Event Recognition with Fragmented Object Tracks

Complete and accurate video tracking is very difficult to achieve in practice due to long occlusions, traffic clutter, shadows and appearance changes. In this paper, we study the feasibility of event recognition when object tracks are fragmented. By changing the lock score threshold controlling track termination, different levels of track fragmentation are generated. The effect on event recognition is revealed by examining the event model match score as a function of lock score threshold. Using a dynamic Bayesian network to model events, it is shown that event recognition actually improves with greater track fragmentation, assuming fragmented tracks for the same object are linked together. The improvement continues up to a point when it is more likely to be offset by other errors such as those caused by frequent object reinitialization. The study is conducted on busy scenes of airplane servicing activities where long tracking gaps occur intermittently

Nonparametric Intensity Priors for Level Set Segmentation of Low Contrast Structures

Segmentation of low contrast objects is an important task in clinical applications like lesion analysis and vascular wall remodeling analysis. Several solutions to low contrast segmentation that exploit high-level information have been previously proposed, such as shape priors and generative models. In this work, we incorporate a priori distributions of intensity and low-level image information into a nonparametric dissimilarity measure that defines a local indicator function for the likelihood of belonging to a foreground object. We then integrate the indicator function into a level set formulation for segmenting low contrast structures. We apply the technique to the clinical problem of positive remodeling of the vessel wall in cardiac CT angiography images. We present results on a dataset of twenty five patient scans, showing improvement over conventional gradient-based level sets.