Search

The development of deep learning models in medical image analysis is majorly limited by the lack of large-sized and well-annotated datasets. Unsupervised learning does not require labels and is more suitable for solving medical image analysis problems. However, most of the current unsupervised learning methods need to be applied to large datasets. To make unsupervised learning applicable to small datasets, we proposed Swin MAE, which is a masked autoencoder with Swin Transformer as its backbone. Even on a dataset of only a few thousand medical images and without using any pre-trained models, Swin MAE is still able to learn useful semantic features purely from images. It can equal or even slightly outperform the supervised model obtained by Swin Transformer trained on ImageNet in terms of the transfer learning results of downstream tasks. The code is publicly available at https://github.com/Zian-Xu/Swin-MAE.

Magnetic Resonance Imaging (MRI) plays an important role in diagnosing the parotid tumor, where accurate segmentation of tumors is highly desired for determining appropriate treatment plans and avoiding unnecessary surgery. However, the task remains nontrivial and challenging due to ambiguous boundaries and various sizes of the tumor, as well as the presence of a large number of anatomical structures around the parotid gland that are similar to the tumor. To overcome these problems, we propose a novel anatomy-aware framework for automatic segmentation of parotid tumors from multimodal MRI. First, a Transformer-based multimodal fusion network PT-Net is proposed in this paper. The encoder of PT-Net extracts and fuses contextual information from three modalities of MRI from coarse to fine, to obtain cross-modality and multi-scale tumor information. The decoder stacks the feature maps of different modalities and calibrates the multimodal information using the channel attention mechanism. Second, considering that the segmentation model is prone to be disturbed by similar anatomical structures and make wrong predictions, we design anatomy-aware loss. By calculating the distance between the activation regions of the prediction segmentation and the ground truth, our loss function forces the model to distinguish similar anatomical structures with the tumor and make correct predictions. Extensive experiments with MRI scans of the parotid tumor showed that our PT-Net achieved higher segmentation accuracy than existing networks. The anatomy-aware loss outperformed state-of-the-art loss functions for parotid tumor segmentation. Our framework can potentially improve the quality of preoperative diagnosis and surgery planning of parotid tumors., Comment: under review

Parotid gland tumor is a common type of head and neck tumor. Segmentation of the parotid glands and tumors by MR images is important for the treatment of parotid gland tumors. However, segmentation of the parotid glands is particularly challenging due to their variable shape and low contrast with surrounding structures. Recently deep learning has developed rapidly, which can handle complex problems. However, most of the current deep learning methods for processing medical images are still based on supervised learning. Compared with natural images, medical images are difficult to acquire and costly to label. Contrastive learning, as an unsupervised learning method, can more effectively utilize unlabeled medical images. In this paper, we used a Transformer-based contrastive learning method and innovatively trained the contrastive learning network with transfer learning. Then, the output model was transferred to the downstream parotid segmentation task, which improved the performance of the parotid segmentation model on the test set. The improved DSC was 89.60%, MPA was 99.36%, MIoU was 85.11%, and HD was 2.98. All four metrics showed significant improvement compared to the results of using a supervised learning model as a pre-trained model for the parotid segmentation network. In addition, we found that the improvement of the segmentation network by the contrastive learning model was mainly in the encoder part, so this paper also tried to build a contrastive learning network for the decoder part and discussed the problems encountered in the process of building.

Lateral two-dimensional (2D) transitional metal dichalcogenide (TMD) heterostructures have recently attracted a wide attention as promising materials for optoelectronic nanodevices. Due to the nanoscale width of lateral heterojunctions, the study of their optical properties is challenging and requires using subwavelength optical characterization techniques. We investigated the photoresponse of a lateral 2D WSe2/MoSe2 heterostructure using tip-enhanced photoluminescence (TEPL) with nanoscale spatial resolution and with picoscale tip-sample distance dependence. We demonstrate the observation of quantum plasmonic effects in 2D heterostructures on a non-metallic substrate, and we report the nano-optical measurements of the lateral 2D TMD heterojunction width of ~ 150 nm and the charge tunneling distance of ~ 20 pm. Controlling the plasmonic tip location allows for both nano-optical imaging and plasmon-induced hot electron injection into the heterostructure. By adjusting the tip-sample distance, we demonstrated the controllability of the hot-electron injection via the competition of two quantum plasmonic photoluminescence (PL) enhancement and quenching mechanisms. The directional charge transport in the depletion region leads to the increased hot electron injection, enhancing the MoSe2 PL signal. The properties of the directional hot-electron injection in the quantum plasmonic regime make the lateral 2D MoSe2/WSe2 heterostructures promising for quantum nanodevices with tunable photoresponse.

A novel crystal configuration of sandwiched S-Mo-Se structure (Janus SMoSe) at the monolayer limit has been synthesized and carefully characterized in this work. By controlled sulfurization of monolayer MoSe2 the top layer of selenium atoms are substituted by sulfur atoms while the bottom selenium layer remains intact. The peculiar structure of this new material is systematically investigated by Raman, photoluminescence and X-ray photoelectron spectroscopy and confirmed by transmission-electron microscopy and time-of-flight secondary ion mass spectrometry. Density-functional theory calculations are performed to better understand the Raman vibration modes and electronic structures of the Janus SMoSe monolayer, which are found to correlate well with corresponding experimental results. Finally, high basal plane hydrogen evolution reaction (HER) activity is discovered for the Janus monolayer and DFT calculation implies that the activity originates from the synergistic effect of the intrinsic defects and structural strain inherent in the Janus structure., Comment: 22 pages, 12 figures

Band gap tuning in two-dimensional transitional metal dichalcogenides (TMDs) is crucial in fabricating new optoelectronic devices. High resolution photoluminescence (PL) microscopy is needed for accurate band gap characterization. We performed tip-enhanced photoluminescence (TEPL) measurements of monolayer MoSe2 with nanoscale spatial resolution, providing an improved characterization of the band gap correlated with the topography compared with the conventional far field spectroscopy. We also observed PL shifts at the edges and investigated the spatial dependence of the TEPL enhancement factors., Comment: 8 pages, 3 figures. Figure 2 update

We report a systematic study of coherent spin precession and spin dephasing in electron-doped monolayer MoS$_2$. Using time-resolved Kerr rotation spectroscopy and applied in-plane magnetic fields, a nanosecond-timescale Larmor spin precession signal commensurate with $g$-factor $|g_0|\simeq 1.86$ is observed in several different MoS$_2$ samples grown by chemical vapor deposition. The dephasing rate of this oscillatory signal increases linearly with magnetic field, suggesting that the coherence arises from a sub-ensemble of localized electron spins having an inhomogeneously-broadened distribution of $g$-factors, $g_0 + \Delta g$. In contrast to $g_0$, $\Delta g$ is sample-dependent and ranges from 0.042 to 0.115., Comment: 5 pages, 4 figures. Accepted and in press, Nano Letters

The recently-discovered monolayer transition metal dichalcogenides (TMDCs) provide a fertile playground to explore new coupled spin-valley physics. Although robust spin and valley degrees of freedom are inferred from polarized photoluminescence (PL) experiments, PL timescales are necessarily constrained by short-lived (3-100ps) electron-hole recombination. Direct probes of spin/valley polarization dynamics of resident carriers in electron (or hole) doped TMDCs, which may persist long after recombination ceases, are at an early stage. Here we directly measure the coupled spin-valley dynamics in electron-doped MoS_2 and WS_2 monolayers using optical Kerr spectroscopy, and unambiguously reveal very long electron spin lifetimes exceeding 3ns at 5K (2-3 orders of magnitude longer than typical exciton recombination times). In contrast with conventional III-V or II-VI semiconductors, spin relaxation accelerates rapidly in small transverse magnetic fields. Supported by a model of coupled spin-valley dynamics, these results indicate a novel mechanism of itinerant electron spin dephasing in the rapidly-fluctuating internal spin-orbit field in TMDCs, driven by fast intervalley scattering. Additionally, a long-lived spin coherence is observed at lower energies, commensurate with localized states. These studies provide crucial insight into the physics underpinning spin and valley dynamics of resident electrons in atomically-thin TMDCs., Comment: 8 pages

Density matrix embedding theory (Phys. Rev. Lett. 109, 186404 (2012)) and density embedding theory ((Phys. Rev. B 89, 035140 (2014)) have recently been introduced for model lattice Hamiltonians and molecular systems. In the present work, the formalism is extended to the ab initio description of infinite systems. An appropriate definition of the impurity Hamiltonian for such systems is presented and demonstrated in cases of 1, 2 and 3 dimensions, using coupled cluster theory as the impurity solver. Additionally, we discuss the challenges related to disentanglement of fragment and bath states. The current approach yields results comparable to coupled cluster calculations of infinite systems even when using a single unit cell as the fragment. The theory is formulated in the basis of Wannier functions but it does not require separate localization of unoccupied bands. The embedding scheme presented here is a promising way of employing highly accurate electronic structure methods for extended systems at a fraction of their original computational cost.