Robust automated organ segmentation is a prerequisite for computer-aided diagnosis (CAD), quantitative imaging analysis and surgical assistance. For high-variability organs such as the pancreas, previous approaches report undesirably low accuracies.
We present a bottom-up approach for pancreas segmentation in abdominal CT scans that is based on a hierarchy of information propagation by classifying image patches at different resolutions; and cascading superpixels. There are four stages: 1) decomposing CT slice images as a set of disjoint boundary-preserving superpixels; 2) computing pancreas class probability maps via dense patch labeling; 3) classifying superpixels by pooling both intensity and probability features to form empirical statistics in cascaded random forest frameworks; and 4) simple connectivity based post-processing. The dense image patch labeling are conducted by: efficient random forest classifier on image histogram, location and texture features; and more expensive (but with better specificity) deep convolutional neural network classification on larger image windows (with more spatial contexts). Evaluation of the approach is performed on a database of 80 manually segmented CT volumes in six-fold cross-validation (CV). Our achieved results are comparable, or better than the state-of-the-art methods (evaluated by leave-one-patient-out), with Dice 70.7% and Jaccard 57.9%. The computational efficiency has been drastically improved in the order of 6~8 minutes, comparing with others of ~10 hours per case. Finally, we implement a multi-atlas label fusion (MALF) approach for pancreas segmentation using the same datasets. Under six-fold CV, our bottom-up segmentation method significantly outperforms its MALF counterpart: (70.7 +/- 13.0%) versus (52.5 +/- 20.8%) in Dice. Deep CNN patch labeling confidences offer more numerical stability, reflected by smaller standard deviations.
We have developed a fully consistent framework for calculations in the Quasiparticle Random Phase Approximation (QRPA) with $NN$ interactions from the Similarity Renormalization Group (SRG) and other unitary transformations of realistic interactions.
The consistency of our calculations, which use the same Hamiltonian to determine the Hartree-Fock-Bogoliubov (HFB) ground states and the residual interaction for QRPA, guarantees an excellent decoupling of spurious strength, without the need for empirical corrections. While work is under way to include SRG-evolved 3N interactions, we presently account for some 3N effects by means of a linearly density-dependent interaction, whose strength is adjusted to reproduce the charge radii of closed-shell nuclei across the whole nuclear chart. As a first application, we perform a survey of the monopole, dipole, and quadrupole response of the calcium isotopic chain and of the underlying single-particle spectra, focusing on how their properties depend on the SRG parameter $lambda$. Unrealistic spin-orbit splittings suggest that spin-orbit terms from the 3N interaction are called for. Nevertheless, our general findings are comparable to results from phenomenological QRPA calculations using Skyrme or Gogny energy density functionals. Potentially interesting phenomena related to low-lying strength warrant more systematic investigations in the future.
Using the ground-state energy of 16-O obtained with the realistic V_UCOM interaction as a test case, we present a comprehensive comparison of different configuration interaction (CI) and coupled-cluster (CC) methods, analyzing the intrinsic advantage
s and limitations of each of the approaches. In particular, we use the importance-truncated (IT) CI and no-core shell model (NCSM) schemes with up to 4-particle-4-hole (4p4h) excitations as well as the size extensive CC methods with a complete treatment of one- and two-body clusters (CCSD) and a non-iterative treatment of connected three-body clusters via the completely renormalized correction to the CCSD energy defining the CR-CC(2,3) approach. We discuss the impact of the center-of-mass contaminations, the choice of the single-particle basis, and size-extensivity on the resulting energies. When the IT-CI and IT-NCSM methods include the 4p4h excitations and when the CC calculations include the 1p1h, 2p2h, and 3p3h clusters, as in the CR-CC(2,3) approach, we observe an excellent agreement among the different methodologies. This shows that despite their individual limitations, the IT-CI, IT-NCSM, and CC methods can provide precise and consistent ab initio nuclear structure predictions. Furthermore, the IT-CI, IT-NCSM, and CC ground-state energy values obtained with 16-O are in good agreement with the experimental value, proving that the V_UCOM two-body interaction allows for a realistic description of binding energies for heavier nuclei and that all of the methods used in this study account for most of the relevant particle correlation effects.
We discuss relations and differences between two methods for the construction of unitarily transformed effective interactions, the Similarity Renormalization Group (SRG) and Unitary Correlation Operator Method (UCOM). The aim of both methods is to co
nstruct a soft phase-shift equivalent effective interaction which is well suited for many-body calculations in limited model spaces. After contrasting the two conceptual frameworks, we establish a formal connection between the initial SRG-generator and the static generators of the UCOM transformation. Furthermore we propose a mapping procedure to extract UCOM correlation functions from the SRG evolution. We compare the effective interactions resulting from the UCOM-transformation and the SRG-evolution on the level of matrix elements, in no-core shell model calculations of light nuclei, and in Hartree-Fock calculations up to 208-Pb. Both interactions exhibit very similar convergence properties in light nuclei but show a different systematic behavior as function of particle number.
We examine the response of closed-shell nuclei using a correlated interaction, derived with the Unitary Correlation Operator Method (UCOM) from the Argonne V18 potential, in second RPA (SRPA) calculations. The same correlated two-body interaction is
used to derive the Hartree-Fock ground state and the SRPA equations. Our results show that the coupling of particle-hole states to higher-order configurations produces sizable effects compared with first-order RPA. A much improved description of the isovector dipole and isoscalar quadrupole resonances is obtained, thanks in part to the more fundamental treatment of the nucleon effective mass offered by SRPA. The present work suggests the prospect of describing giant resonance properties realistically and consistently within extended RPA theories. Self-consistency issues of the present SRPA method and residual three-body effects are pointed out.
We respond to Comment on our recent letter (Phys.Rev.Lett.99:092501,2007) by Dean et al (arXiv:0709.0449).
Lately we have been tackling the problem of describing nuclear collective excitations starting from correlated realistic nucleon-nucleon (NN) interactions. The latter are constructed within the Unitary Correlation Operator Method (UCOM), starting fro
m realistic NN potentials. It has been concluded that first-order RPA with a two-body UCOM interaction is not capable, in general, of reproducing quantitatively the properties of giant resonances (GRs), due to missing higher-order configurations and long-range correlations as well as neglected three-body terms in the Hamiltonian. Here we report results on GRs obtained by employing a UCOM interaction based on the Argonne V18 potential in Second RPA (SRPA) calculations. The same interaction is used to describe the Hartree-Fock (HF) ground state and the residual interactions. We find that the inclusion of second-order configurations -- which effectively dress the underlying HF single-particle states with self-energy insertions -- produces sizable corrections. The effect appears essential for a realistic description of GRs when using the UCOM. We argue that effects of higher than second order should be negligible. Therefore, the UCOM-SRPA emerges as a promising tool for consistent calculations of collective states in closed-shell nuclei. This is an interesting development, given that SRPA can accommodate more physics than RPA (e.g., fragmentation). Remaining discrepancies due to the missing three-body terms and self-consistency issues of the present SRPA model are pointed out.
We propose an importance truncation scheme for the no-core shell model, which enables converged calculations for nuclei well beyond the p-shell. It is based on an a priori measure for the importance of individual basis states constructed by means of
many-body perturbation theory. Only the physically relevant states of the no-core model space are considered, which leads to a dramatic reduction of the basis dimension. We analyze the validity and efficiency of this truncation scheme using different realistic nucleon-nucleon interactions and compare to conventional no-core shell model calculations for 4He and 16O. Then, we present the first converged calculations for the ground state of 40Ca within no-core model spaces including up to 16hbarOmega-excitations using realistic low-momentum interactions. The scheme is universal and can be easily applied to other quantum many-body problems.
