A Comparative Study of Deep Neural Networks for Real-Time Semantic Segmentation during the Transurethral Resection of Bladder Tumors.

Bladder cancer is a common and often fatal disease. Papillary bladder tumors are well detectable using cystoscopic imaging, but small or flat lesions are frequently overlooked by urologists. However, detection accuracy can be improved if the images from the cystoscope are segmented in real time by a deep neural network (DNN). In this paper, we compare eight state-of-the-art DNNs for the semantic segmentation of white-light cystoscopy images: U-Net, UNet++, MA-Net, LinkNet, FPN, PAN, DeepLabv3, and DeepLabv3+. The evaluation includes per-image classification accuracy, per-pixel localization accuracy, prediction speed, and model size. Results show that the best F-score for bladder cancer (91%), the best segmentation map precision (92.91%), and the lowest size (7.93 MB) are also achieved by the PAN model, while the highest speed (6.73 ms) is obtained by DeepLabv3+. These results indicate better tumor localization accuracy than reported in previous studies. It can be concluded that deep neural networks may be extremely useful in the real-time diagnosis and therapy of bladder cancer, and among the eight investigated models, PAN shows the most promising results.

Robust Multiple Importance Sampling with Tsallis φ-Divergences.

Multiple Importance Sampling (MIS) combines the probability density functions (pdf) of several sampling techniques. The combination weights depend on the proportion of samples used for the particular techniques. Weights can be found by optimization of the variance, but this approach is costly and numerically unstable. We show in this paper that MIS can be represented as a divergence problem between the integrand and the pdf, which leads to simpler computations and more robust solutions. The proposed idea is validated with 1D numerical examples and with the illumination problem of computer graphics.

Multiple importance sampling for PET.

This paper proposes the application of multiple importance sampling in fully 3-D positron emission tomography to speed up the iterative reconstruction process. The proposed method combines the results of lines of responses (LOR) driven and voxel driven projections keeping their advantages, like importance sampling, performance and parallel execution on graphics processing units. Voxel driven methods can focus on point like features while LOR driven approaches are efficient in reconstructing homogeneous regions. The theoretical basis of the combination is the application of the mixture of the samples generated by the individual importance sampling methods, emphasizing a particular method where it is better than others. The proposed algorithms are built into the Tera-tomo system.

Averaging and Metropolis iterations for positron emission tomography.

Iterative positron emission tomography (PET) reconstruction computes projections between the voxel space and the lines of response (LOR) space, which are mathematically equivalent to the evaluation of multi-dimensional integrals. The dimension of the integration domain can be very high if scattering needs to be compensated. Monte Carlo (MC) quadrature is a straightforward method to approximate high-dimensional integrals. As the numbers of voxels and LORs can be in the order of hundred millions and the projection also depends on the measured object, the quadratures cannot be precomputed, but Monte Carlo simulation should take place on-the-fly during the iterative reconstruction process. This paper presents modifications of the maximum likelihood, expectation maximization (ML-EM) iteration scheme to reduce the reconstruction error due to the on-the-fly MC approximations of forward and back projections. If the MC sample locations are the same in every iteration step of the ML-EM scheme, then the approximation error will lead to a modified reconstruction result. However, when random estimates are statistically independent in different iteration steps, then the iteration may either diverge or fluctuate around the solution. Our goal is to increase the accuracy and the stability of the iterative solution while keeping the number of random samples and therefore the reconstruction time low. We first analyze the error behavior of ML-EM iteration with on-the-fly MC projections, then propose two solutions: averaging iteration and Metropolis iteration. Averaging iteration averages forward projection estimates during the iteration sequence. Metropolis iteration rejects those forward projection estimates that would compromise the reconstruction and also guarantees the unbiasedness of the tracer density estimate. We demonstrate that these techniques allow a significant reduction of the required number of samples and thus the reconstruction time. The proposed methods are built into the Teratomo system.

Parallel Iteration to the Radiative Transport in Inhomogeneous Media with Bootstrapping.

This paper presents a fast parallel method to solve the radiative transport equation in inhomogeneous participating media. We apply a novel approximation scheme to find a good initial guess for both the direct and scattered components. Then, the initial approximation is used to bootstrap an iterative multiple scattering solver, i.e., we let the iteration concentrate just on the residual problem. This kind of bootstrapping makes the volumetric source approximation more uniform, thus it helps to reduce the discretization artifacts and improves the efficiency of the parallel implementation. The iterative refinement is executed on a face-centered cubic grid. The implementation is based on CUDA and runs on the GPU. For large volumes that do not fit into the GPU memory, we also consider the implementation on a GPU cluster, where the volume is decomposed to blocks according to the available GPU nodes. We show how the communication bottleneck can be avoided in the cluster implementation by not exchanging the boundary conditions in every iteration step. In addition to light photons, we also discuss the generalization of the method to γ-photons that are relevant in medical simulation.