Group visualization of class-discriminative features.

Research explaining the behavior of convolutional neural networks (CNNs) has gained a lot of attention over the past few years. Although many visualization methods have been proposed to explain network predictions, most fail to provide clear correlations between the target output and the features extracted by convolutional layers. In this work, we define a concept, i.e., class-discriminative feature groups, to specify features that are extracted by groups of convolutional kernels correlated with a particular image class. We propose a detection method to detect class-discriminative feature groups and a visualization method to highlight image regions correlated with particular output and to interpret class-discriminative feature groups intuitively. The experiments showed that the proposed method can disentangle features based on image classes and shed light on what feature groups are extracted from which regions of the image. We also applied this method to visualize "lost" features in adversarial samples and features in an image containing a non-class object to demonstrate its ability to debug why the network failed or succeeded.

Hybrid image of three contents.

A hybrid image allows multiple image interpretations to be modulated by the viewing distance. Originally, it can be constructed by combining the low and high spatial frequencies of two different images. The original hybrid image synthesis was limited to similar shapes of source images that were aligned in the edges, e.g., faces with a different expression, to produce an effective double image interpretation. In our previous work, we proposed a noise-inserted method for synthesizing a hybrid image from dissimilar shape images or unaligned images. In this work, we propose a novel method for adding an image to be seen from a middle viewing distance. The middle-frequency (MF) image is extracted by a special bandpass filter, which generates ringing while extracting only specified frequency bands. With this method, the middle frequency should be perceived as a meaningless pattern when viewed from a far distance and close up. A parameter tuning experiment was performed to determine the suitable cutoff frequencies for designing the filter for the MF image. We found that ringings of a suitable size could be used to make the middle frequency less noticeable when seen from far away.

Correction to: Hybrid image of three contents.

In the original publication of this article [1], the Figs. 3 and 4 are not clear enough. They are adjusted the size and showed as below.

High-resolution inventory of mercury emissions from biomass burning in tropical continents during 2001-2017.

Mercury emissions from biomass burning contribute significantly to the atmospheric mercury budget and the interannual variation of mercury concentrations in the troposphere. This study developed a high-resolution (0.1°â€¯× 0.1°) monthly inventory of mercury emissions from biomass burning across five land types in the tropical continents (Central and South America, Africa, and South and Southeast Asia) during 2001-2017. The inventory estimates of mercury emissions from biomass burning are based on the newly released MCD64A1 Version 6 Burned Area data product, satellite and observational data of biomass density, and spatial and temporal variable combustion factors. Results from the inventory demonstrated that during 2001-2017, the average annual mercury emissions from biomass burning in tropical continents was 497 Mg and ranged from 289 Mg to 681 Mg. Forest fires were the largest contributor, accounting for 61% (300 Mg) of the total mercury emissions from biomass burning, followed by fires in woody savanna/shrubland (30%, 151 Mg), savanna/grassland (7%, 35 Mg), peatland (1%, 6 Mg), and cropland (1%, 5 Mg). However, these proportions varied between the continents; in the Americas and Asia, the largest biomass burning emissions came from forest fires, and in Africa the largest emissions were from fires woody savanna/shrubland. Between the three continents, Africa released 41% of the mercury emissions from biomass burning (202 Mg year-1), Asia released 31% (154 Mg year-1), and the Americas released 28% (141 Mg year-1). The total mercury emissions from biomass burning in these tropical continents exhibited strong interannual variations from 2001 to 2017, with peak emissions in March and August to September, and forest fires were the primary land type controlling the interannual variations.

Grading of Atrophic Gastritis is Useful for Risk Stratification in Endoscopic Screening for Gastric Cancer.

OBJECTIVES: In order to screen for gastric cancer effectively, its interval should be set according to the risk. This study aimed to determine whether risk stratification is possible using the data obtained from medical examination or endoscopic findings. METHODS: First, subjects who underwent both cancer screening and medical examination from 2009 to 2015 and underwent cancer screening once more by 2016 were studied. Data such as the lipid profile and history of smoking obtained during the medical examination, and the grade of atrophy and presence of peptic ulcers were studied using multivariate analysis. Next, subjects who underwent cancer screening twice or more between 2009 and 2015 with or without medical examinations were studied to analyze any correlation between the grade of atrophy and cancer occurrence using univariate analysis. In both studies, the status of Helicobacter pylori (HP) infection was determined. RESULTS: In the multivariate analysis, 9378 subjects were included. Aging, advanced atrophy, presence of ulcers, and uric acid levels were identified as risk factors. Among subjects who underwent successful HP eradication therapy, advanced atrophy and aging were observed to be crucial risk factors. In the univariate analysis, there were 12,941 subjects. Gastric cancer occurred more frequently in the more severe atrophy group (P < 0.001). The annual rate of cancer occurrence in the most severe atrophy group was 0.31%, which was approximately thrice as that in the less atrophy group. CONCLUSIONS: Risk stratification was possible based on endoscopic examination alone. The interval should be set depending on each case.

Underlying causes of PM2.5-induced premature mortality and potential health benefits of air pollution control in South and Southeast Asia from 1999 to 2014.

Quantification of spatial and temporal variations in premature mortality attributable to PM2.5 has important implications for air quality control in South and Southeast Asia (SSEA). The number of PM2.5-induced premature deaths during 1999-2014 in SSEA was estimated using an integrated exposure-response model based on 0.01°â€¯× 0.01° satellite-retrieved PM2.5 data, population density, and spatially and temporally variable baseline mortality data. The results showed extremely high premature death rates in North India and Bangladesh. PM2.5-induced premature deaths in SSEA increased with small interannual variations from 1999 to 2014 owing to the interannual variations in PM2.5 concentrations. Moreover, four scenarios on the effects of premature deaths by PM2.5 mitigation efforts based on World Health Organization (WHO) air quality guidelines (AQG) and interim targets (ITs) were investigated for each disease and each country during 1999-2014. Four scenarios based on WHO AQG (10 µg/m3), IT-3 (15 µg/m3), IT-2 (25 µg/m3), and IT-1 (35 µg/m3) resulted in 69.3%, 49.1%, 25.4%, and 12.8% reductions compared to the total reference premature deaths (1256,300), which was calculated using the original PM2.5 datasets. Overall, stroke was the most serious disease associated with air pollution, causing 40% of total premature deaths. Ischemic heart disease was the largest contributor (58%) to the deaths in relatively cleaner air (Scenario 1). The annual rate of change in premature deaths in South Asian countries (India, Bangladesh, and Pakistan) was higher than that in Southeast Asian countries under all scenarios. The results for different scenarios provide insight into the largest health benefits of PM2.5 reduction efforts.

Long-term trends and spatial patterns of PM2.5-induced premature mortality in South and Southeast Asia from 1999 to 2014.

Fine particulate matter (PM2.5) poses a potential threat to human health, including premature mortality under long-term exposure. Based on a long-term series of high-resolution (0.01°×0.01°) satellite-retrieved PM2.5 concentrations, this study estimated the premature mortality attributable to PM2.5 in South and Southeast Asia (SSEA) from 1999 to 2014. Then, the long-term trends and spatial characteristics of PM2.5-induced premature deaths (1999-2014) were analyzed using trend analyses and standard deviation ellipses. Results showed the estimated number of PM2.5-induced average annual premature deaths in SSEA was 1,447,000. The numbers increased from 1,179,400 in 1999 to 1,724,900 in 2014, with a growth rate of 38% and net increase of 545,500. Stroke and ischemic heart disease were the two principal contributors, accounting for 39% and 35% of the total, respectively. High values were concentrated in North India, Bangladesh, East Pakistan, and some metropolitan areas of Southeast Asia. An estimated 991,600 deaths in India was quantified (i.e., ~69% of the total premature deaths in SSEA). The long-term trends (1999-2014) of PM2.5-related premature mortality exhibited consistent incremental tendencies in all countries except Sri Lanka. The findings of this study suggest that strict controls of PM2.5 concentrations in SSEA are urgently required.

Long-term trends and spatial patterns of satellite-retrieved PM2.5 concentrations in South and Southeast Asia from 1999 to 2014.

Fine particulate matter, or PM2.5, is a serious air pollutant and has significant effects on human health, including premature death. Based on a long-term series of satellite-retrieved PM2.5 concentrations, this study analyzed the spatial and temporal characteristics of PM2.5 in South and Southeast Asia (SSEA) from 1999 to 2014 using standard deviation ellipse and trend analyses. A health risk assessment of human exposure to PM2.5 between 1999 and 2014 was then undertaken. The results show that PM2.5 concentrations increased in most areas of SSEA from 1999 to 2014 and exceeded the World Health Organization average annual limit of primary PM2.5 standards. Bangladesh, Pakistan and India experienced average PM2.5 values higher than the total average for SSEA. From 1999 to 2014, the entirety of SSEA exhibited an increased rate of 0.02µg/m3/year on average. Bangladesh and Myanmar witnessed greater incremental rates of PM2.5 than India. Correspondingly, the center of the average regional PM2.5 concentration gradually shifted to the southeast during the study period. The proportion of areas with PM2.5 concentrations exceeding 35µg/m3 increased consistently, and the areas with PM2.5 concentrations below 15µg/m3 decreased continuously. The proportion of the population exposed to high PM2.5 (above 35µg/m3) increased annually. The extent of high-health-risk areas in SSEA expanded in size and extent between 1999 and 2014, particularly in North India, Bangladesh and East Pakistan. Therefore, all of SSEA should receive special attention, and strict controls on PM2.5 concentrations in SSEA countries are urgently required.

Distinguishing the vegetation dynamics induced by anthropogenic factors using vegetation optical depth and AVHRR NDVI: A cross-border study on the Mongolian Plateau.

Distinguishing the vegetation dynamics induced by anthropogenic factors and identifying the major drivers can provide crucial information for designing actionable and practical countermeasures to restore degraded grassland ecosystems. Based on the residual trend (RESTREND) method, this study distinguished the vegetation dynamics induced by anthropogenic factors from the effects of climate variability on the Mongolian Plateau during 1993-2012 using vegetation optical depth (VOD) and normalized difference vegetation index (NDVI), which measure vegetation water content in aboveground biomass and chlorophyll abundance in canopy cover respectively; afterwards, the major drivers within different agricultural zones and socio-institutional periods were identified by integrating agricultural statistics with statistical analysis techniques. The results showed that grasslands in Mongolia and the grazing zone of Inner Mongolia Autonomous Region (IMAR), China underwent a significant human-induced decrease in aboveground biomass during 1993-2012 and 1993-2000 respectively, which was attributable to the rapid growth of livestock densities stimulated by livestock privatization and market factors; by contrast, grasslands in these two regions did not experience a concurrent human-induced reduction in canopy greenness. Besides, the results indicated that grasslands in the grazing zone of IMAR underwent a significant human-induced increase in aboveground biomass since 2000, which was attributable to the reduced grazing pressure induced by China's ecological restoration programs; concurrently, grasslands in this region also experienced a remarkable increase in canopy greenness, however, this increase was found not directly caused by the decreased stocking densities. Furthermore, the results revealed that the farming and semi-grazing/farming zone of IMAR underwent a significant human-induced increase in both aboveground biomass and canopy greenness since 2000, which was attributable to the intensified grain production stimulated by market factors, open grazing regulation and confined feeding popularization. These findings suggest that China's grassland restoration practice has important implications for Mongolia to reverse the severe and continuous grassland degradation in the future.

Satellite-derived mineral mapping and monitoring of weathering, deposition and erosion.

The Earth's surface comprises minerals diagnostic of weathering, deposition and erosion. The first continental-scale mineral maps generated from an imaging satellite with spectral bands designed to measure clays, quartz and other minerals were released in 2012 for Australia. Here we show how these satellite mineral maps improve our understanding of weathering, erosional and depositional processes in the context of changing weather, climate and tectonics. The clay composition map shows how kaolinite has developed over tectonically stable continental crust in response to deep weathering during northwardly migrating tropical conditions from 45 to 10 Ma. The same clay composition map, in combination with one sensitive to water content, enables the discrimination of illite from montmorillonite clays that typically develop in large depositional environments over thin (sinking) continental crust such as the Lake Eyre Basin. Cutting across these clay patterns are sandy deserts that developed <10 Ma and are well mapped using another satellite product sensitive to the particle size of silicate minerals. This product can also be used to measure temporal gains/losses of surface clay caused by periodic wind erosion (dust) and rainfall inundation (flood) events. The accuracy and information content of these satellite mineral maps are validated using published data.

Comparison of global inventories of CO2 emissions from biomass burning during 2002-2011 derived from multiple satellite products.

This study compared five widely used globally gridded biomass burning emissions inventories for the 2002-2011 period (Global Fire Emissions Database 3 (GFED3), Global Fire Emissions Database 4 (GFED4), Global Fire Assimilation System 1.0 (GFAS1.0), Fire INventory from NCAR 1.0 (FINN1.0) and Global Inventory for Chemistry-Climate studies-GFED4 (G-G)). Average annual CO2 emissions range from 6521.3 to 9661.5 Tg year(-1) for five inventories, with extensive amounts in Africa, South America and Southeast Asia. Coefficient of Variation for Southern America, Northern and Southern Africa are 30%, 39% and 48%. Globally, the majority of CO2 emissions are released from savanna burnings, followed by forest and cropland burnings. The largest differences among the five inventories are mainly attributable to the overestimation of CO2 emissions by FINN1.0 in Southeast Asia savanna and cropland burning, and underestimation in Southern Africa savanna and Amazon forest burning. The overestimation in Africa by G-G also contributes to the differences.

High-Resolution Mapping of Biomass Burning Emissions in Three Tropical Regions.

Biomass burning in tropical regions plays a significant role in atmospheric pollution and climate change. This study quantified a comprehensive monthly biomass burning emissions inventory with 1 km high spatial resolution, which included the burning of vegetation, human waste, and fuelwood for 2010 in three tropical regions. The estimations were based on the available burned area product MCD64A1 and statistical data. The total emissions of all gases and aerosols were 17382 Tg of CO2, 719 Tg of CO, 30 Tg of CH4, 29 Tg of NOx, 114 Tg of NMOC (nonmethane organic compounds), 7 Tg of SO2, 10 Tg of NH3, 79 Tg of PM2.5 (particulate matter), 45 Tg of OC (organic carbon), and 6 Tg of BC (black carbon). Taking CO as an example, vegetation burning accounted for 74% (530 Tg) of the total CO emissions, followed by fuelwood combustion and human waste burning. Africa was the biggest emitter (440 Tg), larger than Central and South America (113 Tg) and South and Southeast Asia (166 Tg). We also noticed that the dominant fire types in vegetation burning of these three regions were woody savanna/shrubland, savanna/grassland, and forest, respectively. Although there were some slight overestimations, our results are supported by comparisons with previously published data.

Multiple factors influencing the release of hTERT mRNA from pancreatic cancer cell lines in in vitro culture.

Since telomerase expression is highly prevalent in human cancers, the quantitation of serum/plasma hTERT (human telomerase reverse transcriptase) mRNA levels may be useful for early detection of PCa (pancreatic cancer). To analyse the correspondence between exhTERT (extracellular hTERT) mRNA levels and hTERT expression, we designed a cell culture system to investigate factors modulating the extracellular levels of hTERT mRNA in media conditioned by eight PCa cell lines. We found that the level of exhTERT mRNA was dependent on cell growth rate. MIAPaCa-2, PANC-1, KLM-1 and PK-9 cells expressed high levels of exhTERT mRNA, independent of cell density, whereas proliferating PK-59, BxPC-3 and PK-45H cells released low levels of exhTERT mRNA. The augmented release of mRNA by spontaneous dead MIAPaCa-2 cells was further increased at postconfluence. In Capan-1 cells, low correspondence of marker was also due to RNase secretion. Upon reaching confluence, some PCa cell lines showed down-regulation of hTERT expression. Following cell-cell adhesion, as shown by E-cadherin engagement, PK-59 cells showed levels of extracellular message below the limits of detection, a loss not due to an increase in message degradation. These results suggest that the levels of exhTERT mRNA in the medium of PCa cell lines are altered not only in response to cell growth rate and cell destruction, but are responsive to extracellular cues such as RNases and cell density. A cell-free assay for exhTERT mRNA may therefore not be useful for early detection of PCa.

Advanced hepatocellular carcinoma treated effectively with irinotecan via hepatic arterial infusion followed by proton beam therapy.

We report a 48-year-old man with hepatocellular carcinoma (HCC) treated with hepatic arterial infusion (HAI) chemotherapy followed by proton beam therapy. The HCC lesion in this patient was 88 mm in diameter, with portal vein tumor thrombosis in the right lobe of the liver. He was first treated with 5-fluorouracil, cisplatin, and isovorin, administered by HAI, combined with interferon-alpha, and he was subsequently treated with epirubicin and mitomycin-C administered by HAI. However, no definite efficacy of either of these treatments was observed. Then, after 3 weeks' continuous administration of irinotecan by HAI, the tumor size decreased to 68 mm in diameter. However, 3 months after reduction of the tumor, the tumor had become enlarged to 100 mm in diameter and intrahepatic metastases were prominent. Angiographic findings indicated that the HCC was fed not only from the right hepatic artery but also from the left gastric and right and left subphrenic arteries. After rearrangement of the arteries, and 3 months' continuous HAI chemotherapy with irinotecan, plus hyperthermia, the tumor size had decreased to 50 mm in diameter. The reduction rate of the main tumor according to the Response Evaluation Criteria in Solid Tumors was 43%; therefore, the efficacy of this treatment was judged as a partial response. Two months after reduction of the tumor, the patient's serum alpha-fetoprotein (AFP) level was elevated, and so docetaxel was administered by HAI instead of irinotecan. The liver tumors showed gradual enlargement during the administration of docetaxel, although the AFP level was suppressed. Proton beam therapy was instituted and the liver tumors showed necrosis after this therapy. The patient died of hepatic failure and distant metastases 6 years after the onset of HCC. As far as we know, this is the first case report of HCC treated effectively with irinotecan administered by HAI followed by proton beam therapy in which tumor suppression and the long-term survival of the patient were observed.

The global distribution of pure anorthosite on the Moon.

It has been thought that the lunar highland crust was formed by the crystallization and floatation of plagioclase from a global magma ocean, although the actual generation mechanisms are still debated. The composition of the lunar highland crust is therefore important for understanding the formation of such a magma ocean and the subsequent evolution of the Moon. The Multiband Imager on the Selenological and Engineering Explorer (SELENE) has a high spatial resolution of optimized spectral coverage, which should allow a clear view of the composition of the lunar crust. Here we report the global distribution of rocks of high plagioclase abundance (approaching 100 vol.%), using an unambiguous plagioclase absorption band recorded by the SELENE Multiband Imager. If the upper crust indeed consists of nearly 100 vol.% plagioclase, this is significantly higher than previous estimates of 82-92 vol.% (refs 2, 6, 7), providing a valuable constraint on models of lunar magma ocean evolution.

Lunar radar sounder observations of subsurface layers under the nearside maria of the Moon.

Observations of the subsurface geology of the Moon help advance our understanding of lunar origin and evolution. Radar sounding from the Kaguya spacecraft has revealed subsurface layers at an apparent depth of several hundred meters in nearside maria. Comparison with the surface geology in the Serenitatis basin implies that the prominent echoes are probably from buried regolith layers accumulated during the depositional hiatus of mare basalts. The stratification indicates a tectonic quiescence between 3.55 and 2.84 billion years ago; mare ridges were formed subsequently. The basalts that accumulated during this quiet period have a total thickness of only a few hundred meters. These observations suggest that mascon loading did not produce the tectonics in Serenitatis after 3.55 billion years ago. Global cooling probably dominated the tectonics after 2.84 billion years ago.

Using remote sensing and GIS to detect and monitor land use and land cover change in Dhaka Metropolitan of Bangladesh during 1960-2005.

This paper illustrates the result of land use/cover change in Dhaka Metropolitan of Bangladesh using topographic maps and multi-temporal remotely sensed data from 1960 to 2005. The Maximum likelihood supervised classification technique was used to extract information from satellite data, and post-classification change detection method was employed to detect and monitor land use/cover change. Derived land use/cover maps were further validated by using high resolution images such as SPOT, IRS, IKONOS and field data. The overall accuracy of land cover change maps, generated from Landsat and IRS-1D data, ranged from 85% to 90%. The analysis indicated that the urban expansion of Dhaka Metropolitan resulted in the considerable reduction of wetlands, cultivated land, vegetation and water bodies. The maps showed that between 1960 and 2005 built-up areas increased approximately 15,924 ha, while agricultural land decreased 7,614 ha, vegetation decreased 2,336 ha, wetland/lowland decreased 6,385 ha, and water bodies decreased about 864 ha. The amount of urban land increased from 11% (in 1960) to 344% in 2005. Similarly, the growth of landfill/bare soils category was about 256% in the same period. Much of the city's rapid growth in population has been accommodated in informal settlements with little attempt being made to limit the risk of environmental impairments. The study quantified the patterns of land use/cover change for the last 45 years for Dhaka Metropolitan that forms valuable resources for urban planners and decision makers to devise sustainable land use and environmental planning.

Discrepancy Between ASTER- and MODIS- Derived Land Surface Temperatures: Terrain Effects.

The MODerate resolution Imaging Spectroradiometer (MODIS) and the Advanced Spaceborne Thermal Emission Reflection Radiometer (ASTER) are onboard the same satellite platform NASA TERRA. Both MODIS and ASTER offer routine retrieval of land surface temperatures (LSTs), and the ASTER- and MODIS-retrieved LST products have been used worldwide. Because a large fraction of the earth surface consists of mountainous areas, variations in elevation, terrain slope and aspect angles can cause biases in the retrieved LSTs. However, terrain-induced effects are generally neglected in most satellite retrievals, which may generate discrepancy between ASTER and MODIS LSTs. In this paper, we reported the terrain effects on the LST discrepancy with a case examination over a relief area at the Loess Plateau of China. Results showed that the terrain-induced effects were not major, but nevertheless important for the total LST discrepancy. A large local slope did not necessarily lead to a large LST discrepancy. The angle of emitted radiance was more important than the angle of local slope in generating the LST discrepancy. Specifically, the conventional terrain correction may be unsuitable for densely vegetated areas. The distribution of ASTER-to-MODIS emissivity suggested that the terrain correction was included in the generalized split window (GSW) based approach used to rectify MODIS LSTs. Further study should include the classification-induced uncertainty in emissivity for reliable use of satellite-retrieved LSTs over relief areas.

A case of pancreatic cancer with formation of a mass mimicking alcoholic or autoimmune pancreatitis in a young man. Possibility of diagnosis by hypermethylation of pure pancreatic juice.

CONTEXT: Mass-forming pancreatitis can be divided into two distinct types: alcoholic and autoimmune. There have been some cases of an ambiguous diagnosis although care was taken to differentiate between alcoholic mass-forming pancreatitis, focal type autoimmune pancreatitis and pancreatic cancer. CASE REPORT: We report a case of pancreatic cancer mimicking alcoholic or autoimmune pancreatitis with the formation of a mass in a 32-year-old man with a history of heavy drinking. Although both serum immunoglobulin G and immunoglobulin G4 levels were normal, many serum auto-antibodies, including the antinuclear antibody, were detected. After he stopped drinking, abdominal computed tomography showed a pancreatic head mass 28 mm in diameter with little and weak enhancement in the early and delayed phases, respectively. Endoscopic retrograde cholangiopancreatography showed an obstruction of the main pancreatic duct in the pancreatic head and marked stenosis of the lower common bile duct. Although a percutaneous ultrasound-guided pancreatic biopsy demonstrated no evidence of autoimmune pancreatitis, he was treated with prednisolone to test the efficacy of steroid therapy. However, the pancreatic mass became enlarged after steroid therapy, and he underwent surgery during which the mass was found to be pancreatic cancer. Although the patient was treated with gemcitabine, he died 5 months after surgery. We retrospectively assessed DNA hypermethylation in the patient's pure pancreatic juice obtained on admission. We observed hypermethylation of the cancer-specific gene tissue factor pathway inhibitor 2 (TFPI2). CONCLUSION: This finding suggests that if the DNA hypermethylation of pure pancreatic juice had been assayed before steroid therapy, it would have supported the diagnosis of pancreatic cancer, and steroid therapy could have been avoided.