Noninvasive prediction of CCL2 expression level in high-grade glioma patients.

BACKGROUND: Gliomas are recognized as the most frequent type of malignancies in the central nervous system, and efficacious prognostic indicators are essential to treat patients with gliomas and improve their clinical outcomes. The chemokine (C-C motif) ligand 2 (CCL2) is a promising predictor for glioma malignancy and progression. However, at present, the methods to evaluate CCL2 expression level are invasive and operator-dependent. OBJECTIVE: It was expected to noninvasively predict CCL2 expression levels in malignant glioma tissues by magnetic resonance imaging (MRI)-based radiomics and assess the association between the developed radiomics model and prognostic indicators and related genes. METHODS: MRI-based radiomics was used to predict CCL2 expression level using data obtained from The Cancer Imaging Archive (TCIA) and The Cancer Genome Atlas (TCGA) databases. A support vector machine (SVM)-based radiomics model and a logistic regression (LR)-based radiomics model were used to predict the radiomics score, and its correlation with CCL2 expression level was analyzed. RESULTS: The results revealed that there was an association between CCL2 expression level and the overall survival of cases with gliomas, and bioinformatics correlation analysis showed that CCL2 expression level was highly correlated with disease-related pathways, such as mTOR signaling pathway, cGMP-PKG signaling pathway, and MAPK signaling pathway. Both SVM- and LR-based radiomics data robustly predicted CCL2 expression level, and radiomics scores could also be used to predict the overall survival of patients. Moreover, the high/low radiomics scores were highly correlated with the known glioma-related genes, including CD70, CD27, and PDCD1. CONCLUSION: An MRI-based radiomics model was successfully developed, and its clinical benefits were confirmed, including the prediction of CCL2 expression level and patients' prognosis.

Watt-level all polarization-maintaining femtosecond fiber laser source at 1100 †nm.

We demonstrate a compact watt-level all polarization-maintaining (PM) femtosecond fiber laser source at 1100 nm. The fiber laser source is seeded by an all PM fiber mode-locked laser employing a nonlinear amplifying loop mirror. The seed laser can generate stable pulses at a fundamental repetition rate of 40.71 MHz with a signal-to-noise rate of >100 dB and an integrated relative intensity noise of only ∼0.061%. After two-stage external amplification and pulse compression, an output power of ∼1.47 W (corresponding to a pulse energy of ∼36.1 nJ) and a pulse duration of ∼251 fs are obtained. The 1100 nm femtosecond fiber laser is then employed as the excitation light source for multicolor multi-photon fluorescence microscopy of Chinese hamster ovary (CHO) cells stably expressing red fluorescent proteins.

Behavioral dynamics of neuroprotective macrophage polarization in neuropathic pain observed by GHz femtosecond laser two-photon excitation microscopy.

Macrophage polarization in neurotoxic (M1) or neuroprotective (M2) phenotypes is known to play a significant role in neuropathic pain, but its behavioral dynamics and underlying mechanism remain largely unknown. Two-photon excitation microscopy (2PEM) is a promising functional imaging tool for investigating the mechanism of cellular behavior, as using near-infrared excitation wavelengths is less subjected to light scattering. However, the higher-order photobleaching effect in 2PEM can seriously hamper its applications to long-term live-cell studies. Here, we demonstrate a GHz femtosecond (fs) 2PEM that enables hours-long live-cell imaging of macrophage behavior with reduced higher-order photobleaching effect-by leveraging the repetition rate of fs pulses according to the fluorescence lifetime of fluorophores. Using this new functional 2PEM platform, we measure the polarization characteristics of macrophages, especially the long-term cellular behavior in efferocytosis, unveiling the dynamic mechanism of neuroprotective macrophage polarization in neuropathic pain. These efforts can create new opportunities for understanding long-term cellular dynamic behavior in neuropathic pain, as well as other neurobiological problems, and thus dissecting the underlying complex pathogenesis.

Measurement and Data Correction of Channel Sampling Timing Walk-Off of Photonic Analog-to-Digital Converter in Signal Recovery.

A two-channel, time-wavelength interleaved photonic analog-to-digital converter (PADC) system with a sampling rate of 10.4 GSa/s was established, and a concise method for measuring and data correcting the channel sampling timing walk-off of PADCs for signal recovery was proposed. The measurements show that for the two RF signals of f1 = 100 MHz and f2 = 200 MHz, the channel sampling timing walk-off was 12 sampling periods, which results in an ENOB = -0.1051 bits for the 100 MHz directly synthesized signal, while the ENOB improved up to 4.0136 bits using shift synthesis. In addition, the peak limit method (PLM) and normalization processing were introduced to reduce the impacts of signal peak jitter and power inconsistency between two channels, which further improve the ENOB of the 100 MHz signal up to 4.5668 bits. All signals were analyzed and discussed in both time and frequency domains. The 21.1 GHz signal was also collected and converted using the established two-channel PADC system with the data correction method, combining the PLM, normalization, and shift synthesis, showing that the ENOB increased from the initial -0.9181 to 4.1913 bits, which demonstrates that our method can be effectively used for signal recovery in channel-interleaved PADCs.

Low-noise mode-locking in a GHz repetition rate Tm3+-doped fiber laser.

We demonstrate a GHz repetition rate mode-locked Tm3+-doped fiber laser with low noise. Based on a home-made Tm3+-doped barium gallo-germanate fiber with reduced dispersion, a broad optical spectrum of mode-locking is achieved, and its amplified spontaneous emission quantum-limited timing jitter is largely suppressed. Besides, we carefully investigate the influence of the intracavity pump strength on the noise performance of the mode-locked pulses and find that manipulating the intracavity pump power can be an effective method for optimizing the timing jitter and relative intensity noise (RIN). Particularly, RIN, which originated from the relaxation oscillation, can be effectively suppressed by 33 dB at offset frequencies of >1 MHz. The integrated timing jitter and RIN are only 7.9 fs (10 kHz-10 MHz) and 0.05% (10 Hz-10 MHz), respectively.

Identification of telomere-associated gene signatures to predict prognosis and drug sensitivity in glioma.

OBJECTIVE: Gliomas are a heterogeneous group of brain tumors with distinct biological and clinical properties, leading to significant mortality and morbidity. Emerging evidence shows telomere maintenance has implicated in glioma susceptibility and prognosis. In this study, we comprehensively analyzed gene signatures related to telomere maintenance in glioma and their predictive values for predicting the prognosis and drug sensitivity in glioma. METHODS: We initially identified telomere-related genes differentially expressed between low-grade glioma (LGG) and glioblastoma (GBM) and accordingly developed a risk model by univariate and multivariate Cox analysis to assess the expressions of telomere-related genes across the risk groups. Finally, to assess these genes in immune function the anti-tumor medications often used in the clinical treatment of glioma, we computed immune cell infiltration analysis and drug sensitivity analysis. RESULTS: The consensus clustering analysis identified 20 telomere-related genes which split LGG patients into two distinct subtypes. The patient survival, the expressions of key telomere-related DEGs, and immune cell infiltration significantly differed between Cluster 1 and Cluster 2. The LASSO risk model [riskScore=(0.086)*HOXA7+(0.242)*WEE1+(0.247)*IGF2BP3+(0.052)*DUSP10] showed significant differences regarding the 1-, 3-, 5-year overall survival, immune cell infiltration, and drug sensitivity between high- and low-risk groups. The predictive nomogram constructed to quantify the survival probability of each sample at 1, 3, and 5 years was consistent with the actual patient survival. CONCLUSION: Our comprehensive characterization of telomere-associated gene signatures in glioma reveals their possible roles in the development, tumor microenvironment, and prognosis. The study provides some suggestive relationships between four telomere-related genes (HOXA7, WEE1, IGF2BP3, and DUSP10) and glioma prognosis.

Association of Cumulative Exposure to Cardiovascular Health Behaviors and Factors with the Onset and Progression of Arterial Stiffness.

AIM: This study aims to explore the association of cumulative exposure to cardiovascular health behaviors and factors with the onset and progression of arterial stiffness. METHODS: In this study, 24,110 participants were examined from the Kailuan cohort, of which 11,527 had undergone at least two brachial-ankle pulse wave velocity (baPWV) measurements. The cumulative exposure to cardiovascular health behaviors and factors (cumCVH) was calculated as the sum of the cumCVH scores between two consecutive physical examinations, multiplied by the time interval between the two. A logistic regression model was constructed to evaluate the association of cumCVH with arterial stiffness. Generalized linear regression models were used to analyze how cumCVH affects baPWV progression. Moreover, a Cox proportional hazards regression model was used to analyze the effect of cumCVH on the risk of arterial stiffness. RESULTS: In this study, participants were divided into four groups, according to quartiles of cumCVH exposure levels, namely, quartile 1 (Q1), quartile 2 (Q2), quartile 3 (Q3), and quartile 4 (Q4). Logistic regression analysis showed that compared with the Q1 group, the incidence of arterial stiffness in terms of cumCVH among Q2, Q3, and Q4 groups decreased by 16%, 30%, and 39%, respectively. The results of generalized linear regression showed that compared with the Q1 group, the incidence of arterial stiffness in the Q3 and Q4 groups increased by -25.54 and -29.83, respectively. The results of Cox proportional hazards regression showed that compared with the Q1 group, the incidence of arterial stiffness in cumCVH among Q2, Q3, and Q4 groups decreased by 11%, 19%, and 22%, respectively. Sensitivity analyses showed consistency with the main results. CONCLUSIONS: High cumCVH can delay the progression of arterial stiffness and reduce the risk of developing arterial stiffness.

1200-W all polarization-maintaining fiber GHz-femtosecond-pulse laser with good beam quality.

In this work, we demonstrate a 1200-W average power all polarization-maintaining (PM) fiber ultrafast laser system operating at 1.0 µm. In accordance with the numerical modeling, the PM fiber laser system is designed and it delivers linearly-polarized femtosecond pulses at a 1.39-GHz fundamental repetition rate, with a maximum output power of 1214 W - to the best of our knowledge, the highest average power from all PM fiber ultrafast laser at 1.0 µm to date. The pulse width can be compressed to ∼800 fs with a beam quality of M2 < 1.1. This kilowatt-class all PM fiber laser system is expected to open new potential for high energy pulse generation through temporal coherent combination and laser ablation using GHz burst fs laser.

Effects of Optical Sampling Pulse Power, RF Power, and Electronic Back-End Bandwidth on the Performance of Photonic Analog-to-Digital Converter.

The effects of optical sampling pulse power, RF power, and electronic back-end bandwidth on the performance of time- and wavelength-interleaved photonic analog-to-digital converter (PADC) with eight-channel 41.6 GHz pulses have been experimentally investigated in detail. The effective number of bits (ENOB) and peak-to-peak voltage (Vpp) of converted 10.6 GHz electrical signals were used to characterize the effects. For the 1550.116 nm channel with 5.2 G samples per second, an average pulse power of 0 to -10 dBm input to the photoelectric detector (PD) has been tested. The Vpp increased with increasing pulse power. And the ENOB for pulse power -9~-3 dBm was almost the same and all were greater than four. Meanwhile, the ENOB decreased either when the pulse power was more than -2 dBm due to the saturation of PD or when the pulse power was less than -10 dBm due to the non-ignorable noise relative to the converted weak signal. In addition, RF powers of -10~15 dBm were loaded into the Mach-Zehnder modulator (MZM). The Vpp increased with the increase in RF power, and the ENOB also showed an increasing trend. However, higher RF power can saturate the PD and induce greater nonlinearity in MZM, leading to a decrease in ENOB, while lower RF power will convert weak electrical signals with more noise, also resulting in lower ENOB. In addition, the back-end bandwidths of 0.2~8 GHz were studied in the experiments. The Vpp decreased as the back-end bandwidth decreased from 8 to 3 GHz, and remained nearly constant for the bandwidth between the Nyquist bandwidth and the subsampled RF signal frequency. The ENOB was almost the same and all greater than four for a bandwidth from 3 to 8 GHz, and gradually increased up to 6.5 as the back-end bandwidth decreased from the Nyquist bandwidth to 0.25 GHz. A bandwidth slightly larger than the Nyquist bandwidth was recommended for low costs and without compromising performance. In our experiment, the -3 to -5 dBm average pulse power, about 10 dBm RF power, and 3 GHz back-end bandwidth were recommended to accomplish both a high ENOB more than four and large Vpp. Our research provides a solution for selecting optical sampling pulse power, RF power, and electronic back-end bandwidth to achieve low-cost and high-performance PADC.

Accuracy and depth evaluation of clinical low pass genome sequencing in the detection of mosaic aneuploidies and CNVs.

BACKGROUND: Low-pass genome sequencing (LP GS) has shown distinct advantages over traditional methods for the detection of mosaicism. However, no study has systematically evaluated the accuracy of LP GS in the detection of mosaic aneuploidies and copy number variants (CNVs) in prenatal diagnosis. Moreover, the influence of sequencing depth on mosaicism detection of LP GS has not been fully evaluated. METHODS: To evaluate the accuracy of LP GS in the detection of mosaic aneuploidies and mosaic CNVs, 27 samples with known aneuploidies and CNVs and 1 negative female sample were used to generate 6 simulated samples and 21 virtual samples, each sample contained 9 different mosaic levels. Mosaic levels were simulated by pooling reads or DNA from each positive sample and the negative sample according to a series of percentages (ranging from 3 to 40%). Then, the influence of sequencing depth on LP GS in the detection of mosaic aneuploidies and CNVs was evaluated by downsampling. RESULTS: To evaluate the accuracy of LP GS in the detection of mosaic aneuploidies and CNVs, a comparative analysis of mosaic levels was performed using 6 simulated samples and 21 virtual samples with 35 M million (M) uniquely aligned high-quality reads (UAHRs). For mosaic levels > 30%, the average difference (detected mosaic levels vs. theoretical mosaic levels) of 6 mosaic CNVs in simulated samples was 4.0%, and the average difference (detected mosaic levels vs. mosaic levels of Y chromosome) of 6 mosaic aneuploidies and 15 mosaic CNVs in virtual samples was 2.7%. Furthermore, LP GS had a higher detection rate and accuracy for the detection of mosaic aneuploidies and CNVs of larger sizes, especially mosaic aneuploidies. For depth evaluation, the results of LP GS in downsampling samples were compared with those of LP GS using 35 M UAHRs. The detection sensitivity of LP GS for 6 mosaic aneuploidies and 15 mosaic CNVs in virtual samples increased with UAHR. For mosaic levels > 30%, the total detection sensitivity reached a plateau at 30 M UAHRs. With 30 M UAHRs, the total detection sensitivity was 99.2% for virtual samples. CONCLUSIONS: We demonstrated the accuracy of LP GS in mosaicism detection using simulated data and virtual samples, respectively. Thirty M UAHRs (single-end 35 bp) were optimal for LP GS in the detection of mosaic aneuploidies and most mosaic CNVs larger than 1.48 Mb (Megabases) with mosaic levels > 30%. These results could provide a reference for laboratories that perform clinical LP GS in the detection of mosaic aneuploidies and CNVs.

Ultrafine electro-optical frequency comb based on cascade phase modulation with cyclic frequency shifting.

An ultrafine electro-optical frequency comb (EOFC) with plentiful comb teeth is demonstrated. Adopting a single-frequency fiber laser as a light source, cascade phase modulation based on a sinusoidal signal and a frequency-time transformation (FTT) signal is executed to generate the EOFC with high fineness. Meanwhile, a cyclic fast frequency shifting strategy is introduced to boost the number of comb teeth and the bandwidth of the EOFC. As a result, an EOFC with 12600 comb lines covering a broad bandwidth from -6.3 GHz to 6.3 GHz is established, corresponding to an ultrafine comb space of 1 MHz. Moreover, the power fluctuation of a comb tooth is less than 0.5 dBm. This state-of-the-art EOFC has significant potential in the field of precision spectroscopy.

High-sensitivity hemoglobin detection based on polarization-differential spectrophotometry.

Hemoglobin content is recognized as a momentous and fundamental physiological indicator, especially the precise detection of trace hemoglobin is of great significance for early diagnosis and prevention of tumors, cancer, organic injury, etc. Therefore, high-sensitivity hemoglobin detection is imperative. However, effective detection methods and reliable detection systems are still lacking and remain enormous challenges. Herein, we present a synthetical strategy to break through the existing bottleneck based on polarization-differential spectrophotometry and high-performance single-frequency green fiber laser. Importantly, this framework not only has precisely extracted the two-dimensional information of intensity and polarization during the interaction between laser and hemoglobin, but also has taken advantage of the high monochromaticity and fine directivity in the optimized laser source to reduce the undesirable scattered disturbance. Thus, the hemoglobin detection sensitivity of 7.2 × 10-5 g/L has advanced a hundredfold compared with conventional spectrophotometry, and the responsive dynamic range is close to six orders of magnitude. Results indicate that our technology can realize high-sensitivity detection of trace hemoglobin content, holding promising applications for precision medicine and early diagnosis as an optical direct and fast detection method.

Nature-Inspired Surface Engineering for Efficient Atmospheric Water Harvesting.

Atmospheric water harvesting is a sustainable solution to global water shortage, which requires high efficiency, high durability, low cost, and environmentally friendly water collectors. In this paper, we report a novel water collector design based on a nature-inspired hybrid superhydrophilic/superhydrophobic aluminum surface. The surface is fabricated by combining laser and chemical treatments. We achieve a 163° contrast in contact angles between the superhydrophilic pattern and the superhydrophobic background. Such a unique superhydrophilic/superhydrophobic combination presents a self-pumped mechanism, providing the hybrid collector with highly efficient water harvesting performance. Based on simulations and experimental measurements, the water harvesting rate of the repeating units of the pattern was optimized, and the corresponding hybrid collector achieves a water harvesting rate of 0.85 kg m-2 h-1. Additionally, our hybrid collector also exhibits good stability, flexibility, as well as thermal conductivity and hence shows great potential for practical application.

Performance analysis of two typical greenhouse lettuce production systems: commercial hydroponic production and traditional soil cultivation.

Introduction: Due to the shortage of land and water resource, optimization of systems for production in commercial greenhouses is essential for sustainable vegetable supply. The performance of lettuce productivity and the economic benefit in greenhouses using a soil-based system (SBS) and a hydroponic production system (HPS) were compared in this study. Methods: Experiments were conducted in two identical greenhouses over two growth cycles (G1 and G2). Three treatments of irrigation volumes (S1, S2, and S3) were evaluated for SBS while three treatments of nutrient solution concentration (H1, H2, and H3) were evaluated for HPS; the optimal levels from each system were then compared. Results and discussion: HPS was more sensitive to the effects of environmental temperature than SBS because of higher soil buffer capacity. Compared with SBS, higher yield (more than 134%) and higher water productivity (more than 50%) were observed in HPS. We detected significant increases in ascorbic acid by 28.31% and 16.67% and in soluble sugar by 57.84% and 32.23% during G1 and G2, respectively, compared with SBS. However, nitrate accumulated in HPS-grown lettuce. When the nutrient solution was replaced with fresh water 3 days before harvest, the excess nitrate content of harvested lettuce in HPS was removed. The initial investment and total operating cost in HPS were 21.76 times and 47.09% higher than those in SBS, respectively. Consideration of agronomic, quality, and economic indicators showed an overall optimal performance of the H2 treatment. These findings indicated that, in spite of its higher initial investment and requirement of advanced technology and management, HPS was more profitable than SBS for commercial lettuce production.

Wideband ultra-low intensity noise reduction via joint action of gain saturation and out-of-phase polarization mixing effect from a semiconductor optical amplifier.

In this article, the vector dynamics of semiconductor optical amplifiers (SOAs) are systematically analyzed and developed to explore its mechanism of intensity noise suppression. First, theoretical investigation on the gain saturation effect and carrier dynamics is performed via a vectorial model, and the calculated result unravels desynchronized intensity fluctuations of two orthogonal polarization states. Particularly, it predicts an out-of-phase case, which allows the cancellation of the fluctuations via adding up the orthogonally-polarized components, then establishes a synthetic optical field with stable amplitude and dynamic polarization, and thereby enables a remarkable relative intensity noise (RIN) reduction. Here, we term this approach of RIN suppression as out-of-phase polarization mixing (OPM). To validate the OPM mechanism, we conduct an SOA-mediated noise-suppression experiment based on a reliable single-frequency fiber laser (SFFL) with the presence of relaxation oscillation peak, and subsequently carry out a polarization resolvable measurement. By this means, out-of-phase intensity oscillations with respect to the orthogonal polarization states are clearly demonstrated, and consequently enable a maximum suppression amplitude of >75 dB. Notably, the RIN of 1550-nm SFFL, suppressed by joint action of OPM and gain saturation effect, is dramatically reduced to -160 dB/Hz in a wideband of 0.5 MHz∼10 GHz, and the performance of which is excellent by comparing with the corresponding shot noise limit of -161.9 dB/Hz. The proposal of OPM here not only facilitates us to dissect the vector dynamics of SOA but also offers a promising solution to realize wideband near-shot-noise-limited SFFL.

Clinical diagnosis of genetic disorders at both single-nucleotide and chromosomal levels based on BGISEQ-500 platform.

Most variations in the human genome refer to single-nucleotide variation (SNV), small fragment insertions and deletions, and genomic copy number variation (CNV). Many human diseases including genetic disorders are associated with variations in the genome. These disorders are often difficult to be diagnosed because of their complex clinical conditions, therefore, an effective detection method is needed to facilitate clinical diagnosis and prevent birth defects. With the development of high-throughput sequencing technology, the method of targeted sequence capture chip has been extensively used owing to its high throughput, high accuracy, fast speed, and low cost. In this study, we designed a chip that potentially captured the coding region of 3043 genes associated with 4013 monogenic diseases, with an addition of 148 chromosomal abnormalities that can be identified by targeting specific regions. To assess the efficiency, a strategy of combining the BGISEQ500 sequencing platform with the designed chip was utilized to screen variants in 63 patients. Eventually, 67 disease-associated variants were found, 31 of which were novel. The results of the evaluation test also show that this combined strategy complies with the requirements of clinical testing and has proper clinical application value.

Unveiling the complexity of spatiotemporal soliton molecules in real time.

Observing the dynamics of 3D soliton molecules can hold great opportunities for unveiling the mechanism of molecular complexity and other nonlinear problems. In spite of this fantastic potential, real-time visualization of their dynamics occurring on femtosecond-to-picosecond time scales is still challenging, particularly when high-spatiotemporal-resolution and long-term observation are required. In this work, we observe the real-time speckle-resolved spectral-temporal dynamics of 3D soliton molecules for a long time interval using multispeckle spectral-temporal measurement technology. Diverse real-time dynamics of 3D soliton molecules are captured for the first time, including the speckle-resolved birth, spatiotemporal interaction, and internal vibration of 3D soliton molecules. Further studies show that nonlinear spatiotemporal coupling associated with a large average-chirp gradient over the speckled mode profile plays a significant role in these dynamics. These efforts may shed new light on decomposing the complexity of 3D soliton molecules, and create an analogy between 3D soliton molecules and chemical molecules.

Validation and depth evaluation of low-pass genome sequencing in prenatal diagnosis using 387 amniotic fluid samples.

BACKGROUND: Low-pass genome sequencing (LP GS) is an alternative to chromosomal microarray analysis (CMA). However, validations of LP GS as a prenatal diagnostic test for amniotic fluid are rare. Moreover, sequencing depth of LP GS in prenatal diagnosis has not been evaluated. OBJECTIVE: The diagnostic performance of LP GS was compared with CMA using 375 amniotic fluid samples. Then, sequencing depth was evaluated by downsampling. RESULTS: CMA and LP GS had the same diagnostic yield (8.3%, 31/375). LP GS showed all copy number variations (CNVs) detected by CMA and six additional variant of uncertain significance CNVs (>100 kb) in samples with negative CMA results; CNV size influenced LP GS detection sensitivity. CNV detection was greatly influenced by sequencing depth when the CNV size was small or the CNV was located in the azoospermia factor c (AZFc) region of the Y chromosome. Large CNVs were less affected by sequencing depth and more stably detected. There were 155 CNVs detected by LP GS with at least a 50% reciprocal overlap with CNVs detected by CMA. With 25 M uniquely aligned high-quality reads (UAHRs), the detection sensitivity for the 155 CNVs was 99.14%. LP GS using samples with 25 M UAHRs showed the same performance as LP GS using total UAHRs. Considering the detection sensitivity, cost and interpretation workload, 25 M UAHRs are optimal for detecting most aneuploidies and microdeletions/microduplications. CONCLUSION: LP GS is a promising, robust alternative to CMA in clinical settings. A total of 25 M UAHRs are sufficient for detecting aneuploidies and most microdeletions/microduplications.

Large Scale Visual Food Recognition.

Food recognition plays an important role in food choice and intake, which is essential to the health and well-being of humans. It is thus of importance to the computer vision community, and can further support many food-oriented vision and multimodal tasks, e.g., food detection and segmentation, cross-modal recipe retrieval and generation. Unfortunately, we have witnessed remarkable advancements in generic visual recognition for released large-scale datasets, yet largely lags in the food domain. In this paper, we introduce Food2K, which is the largest food recognition dataset with 2,000 categories and over 1 million images. Compared with existing food recognition datasets, Food2K bypasses them in both categories and images by one order of magnitude, and thus establishes a new challenging benchmark to develop advanced models for food visual representation learning. Furthermore, we propose a deep progressive region enhancement network for food recognition, which mainly consists of two components, namely progressive local feature learning and region feature enhancement. The former adopts improved progressive training to learn diverse and complementary local features, while the latter utilizes self-attention to incorporate richer context with multiple scales into local features for further local feature enhancement. Extensive experiments on Food2K demonstrate the effectiveness of our proposed method. More importantly, we have verified better generalization ability of Food2K in various tasks, including food image recognition, food image retrieval, cross-modal recipe retrieval, food detection and segmentation. Food2K can be further explored to benefit more food-relevant tasks including emerging and more complex ones (e.g., nutritional understanding of food), and the trained models on Food2K can be expected as backbones to improve the performance of more food-relevant tasks. We also hope Food2K can serve as a large scale fine-grained visual recognition benchmark, and contributes to the development of large scale fine-grained visual analysis.

Feature issue introduction: ultrafast optical imaging.

This feature issue of Optics Express collects 20 articles that report the most recent progress of ultrafast optical imaging. This review provides a summary of these articles that cover the spectrum of ultrafast optical imaging, from new technologies to applications.