Source author record

Hiroshi Ishikawa

Hiroshi Ishikawa appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

hep-th Computer Vision Machine Learning eess.IV math.QA

Catalog footprint

What is connected

7works

5topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2021arXiv

Uncertainty guided semi-supervised segmentation of retinal layers in OCT images

Deep convolutional neural networks have shown outstanding performance in medical image segmentation tasks. The usual problem when training supervised deep learning methods is the lack of labeled data which is time-consuming and costly to obtain. In this paper, we propose a novel uncertainty-guided semi-supervised learning based on a student-teacher approach for training the segmentation network using limited labeled samples and a large number of unlabeled images. First, a teacher segmentation model is trained from the labeled samples using Bayesian deep learning. The trained model is used to generate soft segmentation labels and uncertainty maps for the unlabeled set. The student model is then updated using the softly segmented samples and the corresponding pixel-wise confidence of the segmentation quality estimated from the uncertainty of the teacher model using a newly designed loss function. Experimental results on a retinal layer segmentation task show that the proposed method improves the segmentation performance in comparison to the fully supervised approach and is on par with the expert annotator. The proposed semi-supervised segmentation framework is a key contribution and applicable for biomedical image segmentation across various imaging modalities where access to annotated medical images is challenging

preprint2020arXiv

Dueling Deep Q-Network for Unsupervised Inter-frame Eye Movement Correction in Optical Coherence Tomography Volumes

In optical coherence tomography (OCT) volumes of retina, the sequential acquisition of the individual slices makes this modality prone to motion artifacts, misalignments between adjacent slices being the most noticeable. Any distortion in OCT volumes can bias structural analysis and influence the outcome of longitudinal studies. On the other hand, presence of speckle noise that is characteristic of this imaging modality, leads to inaccuracies when traditional registration techniques are employed. Also, the lack of a well-defined ground truth makes supervised deep-learning techniques ill-posed to tackle the problem. In this paper, we tackle these issues by using deep reinforcement learning to correct inter-frame movements in an unsupervised manner. Specifically, we use dueling deep Q-network to train an artificial agent to find the optimal policy, i.e. a sequence of actions, that best improves the alignment by maximizing the sum of reward signals. Instead of relying on the ground-truth of transformation parameters to guide the rewarding system, for the first time, we use a combination of intensity based image similarity metrics. Further, to avoid the agent bias towards speckle noise, we ensure the agent can see retinal layers as part of the interacting environment. For quantitative evaluation, we simulate the eye movement artifacts by applying 2D rigid transformations on individual B-scans. The proposed model achieves an average of 0.985 and 0.914 for normalized mutual information and correlation coefficient, respectively. We also compare our model with elastix intensity based medical image registration approach, where significant improvement is achieved by our model for both noisy and denoised volumes.

preprint2020arXiv

Self-supervised Denoising via Diffeomorphic Template Estimation: Application to Optical Coherence Tomography

Optical Coherence Tomography (OCT) is pervasive in both the research and clinical practice of Ophthalmology. However, OCT images are strongly corrupted by noise, limiting their interpretation. Current OCT denoisers leverage assumptions on noise distributions or generate targets for training deep supervised denoisers via averaging of repeat acquisitions. However, recent self-supervised advances allow the training of deep denoising networks using only repeat acquisitions without clean targets as ground truth, reducing the burden of supervised learning. Despite the clear advantages of self-supervised methods, their use is precluded as OCT shows strong structural deformations even between sequential scans of the same subject due to involuntary eye motion. Further, direct nonlinear alignment of repeats induces correlation of the noise between images. In this paper, we propose a joint diffeomorphic template estimation and denoising framework which enables the use of self-supervised denoising for motion deformed repeat acquisitions, without empirically registering their noise realizations. Strong qualitative and quantitative improvements are achieved in denoising OCT images, with generic utility in any imaging modality amenable to multiple exposures.

preprint2003arXiv

Twisted boundary states in c=1 coset conformal field theories

We study the mutual consistency of twisted boundary conditions in the coset conformal field theory G/H. We calculate the overlap of the twisted boundary states of G/H with the untwisted ones, and show that the twisted boundary states are consistently defined in the diagonal modular invariant. The overlap of the twisted boundary states is expressed by the branching functions of a twisted affine Lie algebra. As a check of our argument, we study the diagonal coset theory so(2n)_1 \oplus so(2n)_1/so(2n)_2, which is equivalent with the orbifold S^1/\Z_2. We construct the boundary states twisted by the automorphisms of the unextended Dynkin diagram of so(2n), and show their mutual consistency by identifying their counterpart in the orbifold. For the triality of so(8), the twisted states of the coset theory correspond to neither the Neumann nor the Dirichlet boundary states of the orbifold and yield the conformal boundary states that preserve only the Virasoro algebra.

preprint2003arXiv

Twisted boundary states in Kazama-Suzuki models

We construct Cardy states in the Kazama-Suzuki model G/H x U(1), which satisfy the boundary condition twisted by the automorphisms of the coset theory. We classify all the automorphisms of G/H x U(1) induced from those of the G theory. The automorphism group contains at least a Z_2 as a subgroup corresponding to the charge conjugation. We show that in several models there exist extra elements other than the charge conjugation and that the automorphism group can be larger than Z_2. We give the explicit form of the twisted Cardy states which are associated with the non-trivial automorphisms. It is shown that the resulting states preserve the N=2 superconformal algebra. As an illustration of our construction, we give a detailed study for two hermitian symmetric space models SU(4)/SU(2) x SU(2) x U(1) and SO(8)/SO(6) x U(1) both at level one. We also study the action of the level-rank duality on the Cardy states and find the relation with the exceptional Cardy states originated from a conformal embedding.

preprint2002arXiv

Novel construction of boundary states in coset conformal field theories

We develop a systematic method to solve the Cardy condition for the coset conformal field theory G/H. The problem is equivalent to finding a non-negative integer valued matrix representation (NIM-rep) of the fusion algebra. Based on the relation of the G/H theory with the tensor product theory G x H, we give a map from NIM-reps of G x H to those of G/H. Our map provides a large class of NIM-reps in coset theories. In particular, we give some examples of NIM-reps not factorizable into the G and the H sectors. The action of the simple currents on NIM-reps plays an essential role in our construction. As an illustration of our procedure, we consider the diagonal coset SU(2)_5 x SU(2)_3 /SU(2)_8 to obtain a new NIM-rep based on the conformal embedding su(2)_3 \oplus su(2)_8 \subset sp(6)_1.

preprint1994arXiv

Non-linear Structures in Non-critical NSR String

We investigate the Ward identities of the $\W_{\infty}$ symmetry in the super-Liouville theory coupled to the super-conformal matter of central charge ${\hat c}_M = 1-2(p-q)^2 /pq$. The theory is classified into two chiralities. For the positive chirality, all gravitationally dressed scaling operators are generated from the $q-1$ gravitational primaries by acting one of the ring generators in the R-sector on them repeatedly. After fixing the normalizations of the dressed scaling operators, we find that the Ward identities are expressed in the form of the {\it usual} $\W_q$ algebra constraints as in the bosonic case: $\W^{(k+1)}_n τ=0$, $(k=1,\cdots,q-1 ;~ n \in {\bf Z}_{\geq 1-k})$, where the equations for even and odd $n$ come from the currents in the NS- and the R-sector respectively. The non-linear terms come from the anomalous contributions at the boundaries of moduli space. The negative chirality is defined by interchanging the roles of $p$ and $q$. Then we get the $\W_p $ algebra constraints.