Search

Detecting and monitoring dental plaque is an important issue in research and clinical practice. In this context, new digital imaging methods that permit permanent documentation of the clinical findings could be promising tools. The aim of the study was therefore to investigate whether disclosed plaque can be reliably visualised on 2D and 3D images captured with digital intraoral imaging devices. Clinical examination was the reference method. Twenty subjects (27.5±1.2 years) were included and plaque was measured at three different stages: habitual plaque (T1), after 72 h without oral hygiene (T2) and after a subsequent habitual brushing exercise (T3). At each time point, plaque was disclosed followed by the clinical examination and capturing the 2D and 3D images (intraoral-camera CS 1500 and intraoral-scanner CS 3600; Carestream Dental, Germany). Plaque amounts were recorded on oral and vestibular surfaces of the Ramfjord-teeth (16, 21, 24, 36, 41, 44) using the Rustogi-modified-Navy-Plaque-Index (RMNPI) and expressed as percentage of plaque-containing RMNPI areas of all RMNPI areas. At T1, percentages (mean±SD) obtained from the clinical examination, 2D and 3D images were 62.2±10.6, 65.1±10.0 and 64.4±10.6 resp. increasing to 76.9±8.0, 77.9±8.6 and 77.5±9.4 resp. at T2. After toothbrushing (T3), values decreased to 56.3±11.1, 58.2±12.1 and 61.2±10.8 resp. All methods were able to show statistically significant changes in plaque amounts at the different time points with in part statistically significant but minor differences between them. The Bland-Altmann analysis revealed a good agreement between values from both 2D and 3D images with the clinical examination. The agreement of the scores obtained with the both image-based methods for the single RMNPI areas with the clinical examination was mainly classified as substantial to almost perfect. Amounts of plaque can be reliably detected and monitored on 2D images from an intraoral camera and on 3D images from an intraoral scanner. [ABSTRACT FROM AUTHOR]

Glossopharyngeal neuralgia is a rare condition with neuralgic sharp pain in the pharyngeal and auricular region. Classical glossopharyngeal neuralgia is caused by neurovascular compression at the root entry zone of the nerve. Regarding the rare occurrence of glossopharyngeal neuralgia, we report clinical data and magnetic resonance imaging (MRI) findings in a case series of 19 patients, of whom 18 underwent surgery. Two patients additionally suffered from trigeminal neuralgia and three from additional symptomatic vagal nerve compression. In all patients, ipsilateral neurovascular compression syndrome of the IX cranial nerve could be shown by high-resolution MRI and image processing, which was confirmed intraoperatively. Additional neurovascular compression of the V cranial nerve was shown in patients suffering from trigeminal neuralgia. Vagal nerve neurovascular compression could be seen in all patients during surgery. Sixteen patients were completely pain free after surgery without need of anticonvulsant treatment. As a consequence of the operation, two patients suffered from transient cerebrospinal fluid hypersecretion as a reaction to Teflon implants. One patient suffered postoperatively from deep vein thrombosis and pulmonary embolism. Six patients showed transient cranial nerve dysfunctions (difficulties in swallowing, vocal cord paresis), but all recovered within 1 week. One patient complained of a gnawing and burning pain in the cervical area. Microvascular decompression is a second-line treatment after failure of standard medical treatment with high success in glossopharyngeal neuralgia. High-resolution MRI and 3D visualization of the brainstem and accompanying vessels as well as the cranial nerves is helpful in identifying neurovascular compression before microvascular decompression procedure. [ABSTRACT FROM AUTHOR]

The purpose of this study was to determine if micrographs produced with a colour charged couple device (CCD) camera were comparable in quality to micrographs obtained with a 35-mm colour film camera. The Axiocam (Zeiss, Jena, Germany) CCD camera produced images of 3900 × 3090 pixels.Each picture in a series of 50 pairs of film and digital micrographs was rated independently by each of three observers on the categories of colour, detail and overall impression. A total of 28 of the 50 pairs of micrographs were discarded due to colour differences.Digital micrographs were rated lower on colour and higher on detail compared to film micrographs. The film and digital micrographs were rated equally on overall impression. The difference in colour between film and digital micrographs can be compensated for using image processing software on the digital image. However, such manipulation was not carried out as part of the present study. The study originally included 50 digital micrographs and 50 film micrographs. The micrographs were paired and 22 of the 50 pairs of micrographs were assessed. Thus 28 pairs were not evaluated due to colour differences in the non-manipulated pictures. The interobserver difference on colour in the remaining 22 pairs given as mean square ± standard error of the mean was− 0.27 ± 0.10, 0.62 ± 0.10 and 0.90 ± 0.11. The interobserver difference on detail was− 0.48 ± 0.10, 0.95 ± 0.10 and 1.43 ± 0.10. The interobserver difference on overall impression was− 0.70 ± 0.10, 0.57 ± 0.10 and 1.27 ± 0.10. All the results differed significantly (p < 0.001).Digital micrographs are sufficient in detail, colour and overall impression to replace the traditional 35-mm film recordings for use in histopathology. [ABSTRACT FROM AUTHOR]

Presents an approach that combines the advantages of classification with linear spectral unmixing. Objectives of hyperspectral image analysis in the urban environment; Iterative linear spectral unmixing approach based on pixel-oriented endmember selection; Analysis of airborne hyperspectral data of the DAIS 7915 instrument for automated area-wide differentiation of urban surface cover types in Dresden, Germany.

Traditional approaches to modeling and processing discrete pixels are mainly based on image features or model optimization. These methods often result in excessive shrinkage or expansion of the restored pixel region, inhibiting accurate recovery of the target pixel region shape. This paper proposes a simultaneous source and mask-images optimization model based on skeleton divergence that overcomes these problems. In the proposed model, first, the edge of the entire discrete pixel region is extracted through bilateral filtering. Then, edge information and Delaunay triangulation are used to optimize the entire discrete pixel region. The skeleton is optimized with the skeleton as the local optimization center and the source and mask images are simultaneously optimized through edge guidance. The technique for order of preference by similarity to ideal solution (TOPSIS) and point-cloud regularization verification are subsequently employed to provide the optimal merging strategy and reduce cumulative error. In the regularization verification stage, the model is iteratively simplified via incremental and hierarchical clustering, so that point-cloud sampling is concentrated in the high-curvature region. The results of experiments conducted using the moving-target region in the RGB-depth (RGB-D) data (Technical University of Munich, Germany) indicate that the proposed algorithm is more accurate and suitable for image processing than existing high-performance algorithms. [ABSTRACT FROM AUTHOR]

Conventional ultrasound (US) is the first-line imaging method for abdominal pathologies, but its diagnostic accuracy is operator-dependent, and data storage is usually limited to two-dimensional images. A novel tomographic US system (Curefab CS, Munich, Germany) processes imaging data combined with three-dimensional spatial information using a magnetic field tracking. This enables standardized image presentation in axial planes and a review of the entire examination. The applicability and diagnostic performance of this tomographic US approach was analyzed in an abdominal setting using conventional US as reference. Tomographic US data were successfully compiled in all subjects of a training cohort (20 healthy volunteers) and in 50 patients with abdominal lesions. Image quality (35% and 79% for training and patient cohort respectively) and completeness of organ visualization (45% and 44%) were frequently impaired in tomographic US compared to conventional US. Conventional and tomographic US showed good agreement for measurement of organ sizes in the training cohort (right liver lobe and both kidneys with a median deviation of 5%). In the patient cohort, tomographic US identified 57 of 74 hepatic or renal lesions detected by conventional ultrasound (sensitivity 77%). In conclusion, this study illustrates the diagnostic potential of abdominal tomographic US, but current significant limitations of the tomographic ultrasound device demand further technical improvements before this and comparable approaches can be implemented in clinical practice. [ABSTRACT FROM AUTHOR]

Visibility observations and accurate forecasts are essential in meteorology, requiring a dense network of observation stations. This paper investigates image processing techniques for object detection and visibility determination using static cameras. It proposes a comprehensive method that includes image preprocessing, landmark identification, and visibility estimation, mirroring the observation process of professional meteorological observers. This study validates the visibility observation procedure using the k-nearest neighbors machine learning method across six locations, including four in the Czech Republic, one in the USA, and one in Germany. By comparing our results with professional observations, the paper demonstrates the suitability of the proposed method for operational application, particularly in foggy and low visibility conditions. This versatile method holds potential for adoption by meteorological services worldwide. [ABSTRACT FROM AUTHOR]

Coastal waters are one of the most vulnerable resources that require comprehensive investigation in space and time. One of the key factors for effective coastal monitoring is the use of remote sensing technologies. Since the Coastal Zone Color Scanner (CZCS) in 1978, a long list of space-borne missions had been successfully launched. However, those missions are limited to coastal waters applications. Despite a large number of missions, the existing systems are still facing similar challenges as four decades ago. Spatial and spectral data reconstruction and recovery a high resolution (HR) imagery data from a low resolution (LR) imaging is a challenging a task in many applications. The most promising technique in the field of digital image processing is known as Super Resolution (SR). Many techniques focus on reconstructing information at the sub-pixel level and dividing the original LR space into pixels corresponding to the HR space. Other methods assume that a series of LR images (in time) of a scene scanned from different perspectives (affine) will provide SR. Alternative methods use different data sources and proper image algorithms. In most cases, SR methods will perform a learning process in which the system will try to identify the inherent redundancy in the natural data in order to retrieve HR information from LR based on a spatial correlation between the original images. The learning process can be significantly efficient by using the Convolutional Neural Network (CNN). CNN submit to training through a large dataset that preserves scene's characteristics. The flexibility afforded by CNN is learning nonlinear relationships when reconstructing a spatial characteristic from an LR image to HR image. The main aim of this study is to identify spectral features related to the coastal water and inland water variations at different spatial and temporal scale and integrate them with a multi-scale information system. The main objectives of the study are developing of spatial-temporal-spectral fusion approach for multi-source data collected from the same geographical site; creating a new method for single image reconstruction from non-complementary information scene. The proposed method measures HR given LR by a downscaling process by turning HR into an LR. The deterministic process calculated using a Gaussian filter and by a photographic-focused distribution function. The correlation coefficient (at the LR-pixel level) used as an inverse ratio to upscaling. The proposed architecture is based on a three-convolutional network. In the first stage, the convolution is directly applied to the LR data, and then another sub-pixel convolution layer is subtracted to generate SR data from LR data through an upscaling process. This study performed in two sites, 1) a training site in Israel, 2) a test site in Germany. The training site is shallow seawaters around Oren River, Israel and the test site is Alfsee inland water in Germany. The results in both sites are SR imagery with full Sentinel 2 spectral resolution and spatial resolution of 0.3 m. [ABSTRACT FROM AUTHOR]

Objective. To monitor size and shape changes of bone erosions and changes in BMD in the vicinity of the erosion and in the periarticular trabecular compartment of patients with RA using high-resolution peripheral quantitative CT (HR-pQCT) imaging and to compare an automated three-dimensional (3D) image processing technique with manual measurements of erosion width and depth. Methods. The shape of 40 bone erosions and composition of bone around the erosions were analysed in the MCP joints of 22 RA patients both manually and by semi-automated 3D image processing at two different time points. Periosteal segmentation was performed using volume growing and morphological operations. Image registration was applied for transfer of baseline segmentations to follow-up datasets. Results. Eight erosions decreased in size, 6 increased and 28 remained stable. Increasing erosions were more spherical and smaller at baseline compared with decreasing or stable erosions. BMD in the vicinity of shrinking erosions increased, while it decreased next to expanding erosions. There was moderate agreement in the determination of erosion volume between semi-automated and manual measurements, but agreement was poor when assessing changes in volume over time. Conclusion. Longitudinal changes in erosion size and shape and of BMD in the vicinity of an erosion can be measured. BMD changes are associated with progression and regression of erosions. However, the semi-automated and manual approaches did not classify longitudinal changes of erosion volume in the same way. Further research is necessary to define the nature of these differences. [ABSTRACT FROM AUTHOR]

A new semi-automated method is proposed in this study that detects joint traces from digital images and calculates the length distribution through three-dimensional data structuring. The method comprises detecting the pixels corresponding to the joint trace in digital images and calculating the length distribution through three-dimensional data structuring of the pixels detected as the joint trace. Semantic segmentation based on deep learning techniques is applied to the detection of joint trace pixels to overcome the limitations of rule-based image processing techniques. For training a deep learning network, various rock joint images were sampled and labeled, following which the data sets for training, validation, and testing were prepared through classification and augmentation. Using the prepared datasets, the performance of the classifiers based on widely used semantic segmentation networks, namely U-Net (University of Freiburg, Germany) network and DeepLabV3+ (Google, USA) network, was compared. The classifier based on the DeepLabV3+ network was ultimately selected and applied. The trained classifier successfully detected pixels corresponding to the joint trace in images of both flat and uneven rock surfaces, indicating that the deep learning technique could be applied effectively to joint trace detection from digital images. Further, a data structuring technique is proposed to calculate the joint trace length using pixelwise data detected through the trained classifier. This technique uses point cloud data obtained by photogrammetry, and comprises two-dimensional thinning and segmentation, three-dimensional projection, segmentation, and segment linking. The entire proposed method, from detecting the joint trace pixels to three-dimensional data structuring, was verified by applying it to both simple flat and uneven rock surface models. • New technique for joint trace detection using deep learning network is proposed. • New data structuring technique for pixelwise data is proposed. • Semi-automatic calculation of the trace length is possible using the two techniques. • Proposed techniques is verified by applying it to both flat and uneven rock surfaces. [ABSTRACT FROM AUTHOR]

SUMMARY Passive seismic interferometry is a new promising methodology for seismic exploration. Interferometry allows information about the subsurface structure to be extracted from ambient seismic noise. In this study, we apply the cross-correlation technique to approximately 25 hr of recordings of ambient seismic noise at the Ketzin experimental CO2 storage site, Germany. Common source gathers were generated from the ambient noise for all available receivers along two seismic lines by cross-correlation of noise records. This methodology isolates the interstation Green's functions that can be directly compared to active source gathers. We show that the retrieved response includes surface waves, refracted waves and reflected waves. We use the dispersive behaviour of the retrieved surface waves to infer geological properties in the shallow subsurface and perform passive seismic imaging of the subsurface structure by processing the retrieved reflected waves. [ABSTRACT FROM AUTHOR]

We describe a method to determine the daytime visibility at a station that is located at the University of Karlsruhe, Germany. A comparison of visibilities determined by applying our new method and visibilities observed with standard methods at a nearby station of the German Weather Service (DWD) demonstrate the successful operation of our method. The highest differences occur for very low and very high visibilities. For visibilities >10km, the correlation between visibilities of both methods is 0.73. Additionally, at low sun zenith angles of <30°, an overestimation by the visibility determined with digital photographs (VISIDIP) method of up to 20% has been observed. Partially, the bias could be explained by the different measurement heights and different local conditions which have an effect especially under certain meteorological conditions as fog or precipitation events. In contrast to human observations, the new method presented in this feasibility study is completely automatic and objective. [Copyright &y& Elsevier]

We present here a new method to predict cloud concentration five minutes in advance from all-sky images using the Artificial Neural Networks (ANN). An autoregressive neural network with backpropagation (Ar-BP) was created and trained with four years of all-sky images as inputs. The pictures were taken with a hemispheric sky imager fixed on the roof at the Institute of Meteorology and Climatology (IMUK) of the Leibniz Universität Hannover, Hannover, Germany. Firstly, a statistical method is presented to obtain key information of the pictures. Secondly, a new image-processing algorithm is suggested to optimize the cloud detection process starting with the Haze Index. Finally, the cloud concentration five minutes in advance at the IMUK is forecasted using machine learning methods. A persistence model forecast to provide a reference for comparison was generated. The results are quantified in terms of the root mean square error (RMSE) and the mean absolute error (MAE). The new algorithm reduced both the RMSE and the MAE of the prediction by approximately 30% compared to the reference persistence model under diverse cloud conditions. The new algorithm could be used as a tool for the stable maintenance of the network for the transmission system operators, i.e., the primary control reserve (within 30 seconds) and the secondary control reserve (within 5 minutes). [ABSTRACT FROM AUTHOR]

Background: Scoliosis is defined as a curvature of the spine in the coronal plane, with its idiopathic form being the most frequently encountered. Diagnosis is based on spinal X-Ray imaging, performing a Cobb angle measurement, which represents the angle calculated between two perpendicular lines traced at the superior margin of the uppermost involved vertebra, and the inferior margin of the lowest affected vertebra. A Cobb angle greater than 10 degrees is considered as positive evidence of spinal deformity and confirms the diagnosis. The traditional Cobb angle measurements are time-consuming and prone to human errors; therefore, an automated measuring algorithm could reduce the diagnostic interval, and hypothetically improve measurement accuracy. Objective: The aim of this study is to investigate and test the application of artificial intelligence (AI), and its existing algorithms, for automated Cobb angle measurement and scoliosis assessment. Material and methods: Radiographic images of 197 patients have been obtained from the database of a private clinic from Targu-Mures, spanning the period between 30/11/2023 and 20/02/2024. The acquired images underwent processing using the AI Open-Source Software "Cobb Angle Calculator". Additionally, a manual calculation was performed using the "RadiAnt DICOM Viewer" software to ensure a comparison between automated and manual methods. Statistical analysis was carried out utilizing the GraphPad InStat software to statistically assess differences between the automated and the manually calculated measurements. Results: Out of the 197 analyzed X-Ray images, the manual measurement of the Cobb angle led to a diagnosis of scoliosis in 170 patients. The measurements were conducted three times, in order to minimize errors. Among these, 52 patients were diagnosed with thoracic scoliosis, 51 with lumbar deformities, and 67 with combined thoraco-lumbar curvatures. The roentgenograms were then processed with the AI software "Cobb Angle Calculator", and the outcomes compared with the ones calculated manually. Statistical analysis demonstrated that the difference in accuracy of the two methods is extremely significant (p value=0.0002), thus rejecting the null hypothesis. Subsequently, each X-ray image underwent a contrast-enhancing process to improve visualization of the skeletal structures, and the results were compared a second time. The obtained p value of 0.3019, considered not statistically significant, confirmed the null hypothesis, therefore proving the valuable aid of AI in Cobb angle quantification as an alternative to manual evaluation. Conclusions: Automated Cobb angle measurement, being at the beginning of its era, lacks sufficient accuracy when employed without a specific image contrast-enhancing procedure, hence displaying high probability of scoliosis misdiagnosis. Integrating an image contrast-enhancing process is a pivotal step forward in measurement automation, and consequently offers a crucial alternative, and possibly future substitution, to the available standard techniques. [ABSTRACT FROM AUTHOR]

Purpose: Seizures related to tumor growth are common in glioma patients, especially in low-grade glioma patients this is often the first tumor manifestation. We hypothesize that there are associations between preoperative seizures and morphologic features (e.g., tumor size, location) and histogram features in patients with glioblastoma (GB). Methods: Retrospectively, 160 consecutive patients with initial diagnosis and surgery of GB (WHO IV) and preoperative MRI were analyzed. Preoperative MRI sequences were co-registered (T2-FLAIR, T1-contrast, DTI) and tumors were segmented by a neuroradiologist using the software ITK-snap blinded to the clinical data. Tumor volume (FLAIR, T1-contrast) and histogram analyses of ADC- and FA-maps were recorded in the contrast enhancing tumor part (CET) and the non-enhancing peritumoral edema (FLAIR). Location was determined after co-registration of the data with an atlas. Permutation-based multiple-testing adjusted t statistics were calculated to compare imaging variables between patients with and without seizures. Results: Patients with seizures showed significantly smaller tumors (CET, adj. p = 0.029) than patients without preoperative seizures. Less seizures were observed in patients with tumor location in the right cingulate gyrus (adj. p = 0.048) and in the right caudate nucleus (adj. p = 0.009). Significant differences of histogram analyses of FA in the contrast enhancing tumor part were observed between patients with and without seizures considering also tumor location and size. Conclusion: Preoperative seizures in GB patients are associated with lower preoperative tumor volume. The different histogram analyses suggest that there might be microstructural differences in the contrast enhancing tumor part of patients with seizures measured by fractional anisotropy. Higher variance of GB presenting without seizures might indicate a more aggressive growth of these tumors. [ABSTRACT FROM AUTHOR]

Objective. To develop a precise three-dimensional (3D) segmentation technique for bone erosions in high-resolution peripheral quantitative CT (HR-pQCT) datasets to measure their volume, surface area and shape parameters. Assessment of bone erosions in patients with RA is important for diagnosis and evaluation of treatment efficacy. HR-pQCT allows quantifying periarticular bone loss in arthritis.Methods. HR-pQCT scans with a spatial resolution of about 120 µm of the second to fourth metacarpophalangeal joints were acquired in patients with RA. Erosions were identified by placing a seed point in each of them. After applying 3D segmentation, the volume, surface area and sphericity of erosions were calculated. Results were compared with an approximation method using manual measurements. Intra- and interoperator precision analysis was performed for both the 3D segmentation and the manual technique.Results. Forty-three erosions were assessed in 18 datasets. Intra- and interoperator precisions (RMSCV/RMSSD) for erosion volume were 5.66%/0.49 mm3 and 7.76%/0.76 mm3, respectively. The correlation between manual measurements and their simulation using segmentation volumes was r = 0.87. Precision errors for the manual method were 15.39% and 0.36 mm3, respectively.Conclusion. We developed a new precise 3D segmentation technique for quantification of bone erosions in HR-pQCT datasets that correlates to the volume, shape and surface area of the erosion. The technique allows fast and effective measurement of the erosion size and could therefore be helpful for rapid and quantitative assessment of erosion size. [ABSTRACT FROM PUBLISHER]

Since 1980, severe thunderstorms have caused nearly USD 3 billion in damages annually inEurope. Hailstorms account for a significant portion of thunderstorm-related losses, yet theclimatological aspects of large hail are not well understood. While the European SevereWeather Database (ESWD) contains the most comprehensive archive of hail observations inEurope, spatial and temporal inhomogeneities and discontinuities preclude the database frombeing used to create a representative long-term hail climatology. We therefore propose amethod to generate a hail event set for Central Europe by: 1) objectively identifying recenthail swaths from high-resolution radar data, and 2) combining the identified hail swathproperties with a probabilistic convective hazard model and ESWD hail observations tosimulate individual hailstorms. Such a hail event set will facilitate the evaluation of theclimatological aspects of large hail and time-dependent changes in the frequency ofhailstorms. In addition, the hail event set will be combined with exposure and vulnerabilitydata to produce detailed loss estimates and better understand the risk of large hail inEurope. Hail swaths occurring in Germany over a 4-year period (2015–2018) were identifiedusing the vertically integrated ice (VII) product from the Deutscher Wetterdienst (DWD).Specifically, we isolated continuous regions where 6-hour accumulated VII satisfied aminimum intensity (25 kg m−3) criterion, and then used Python image processing to generatehail swath ellipses and analyze their spatial characteristics. Next, for each 6-hour intervalduring the 1979–2018 period, we applied the convective hazard model to ERA-Interimreanalysis data to simulate the occurrence/non-occurrence of hail in each reanalysis grid cell.If hail was simulated in a given grid cell, we also predicted the number of hail swaths in thegrid cell during the 6-hour period. For each hail swath, we simulated the length,area, and orientation based on the spatial properties of the previously identified hailswaths. In addition, we simulated the maximum hail diameter based on the statisticaldistributions of observed hail diameter under various instability and deep-layer shearscenarios. Results from the VII hail detection algorithm suggest that large hail in Germany occursprimarily after 12 UTC, with a maximum frequency during the afternoon and early evening(12–18 UTC). While the majority of identified hail swaths affected areas less than 100km2, hail swaths exceeding 500 km2 were not especially uncommon. Hail swathsgenerally followed trajectories between 225˚ (southwest-to-northeast) and 315˚ (northwest-to-southeast) and were typically associated with right-moving thunderstorm cellswhen the synoptic-scale environment was moderately-to-strongly sheared. The simulated hailswaths exhibit a meridional gradient in the predicted frequency of large hail, with thelowest frequencies near the North and Baltic Seas, and the highest frequencies overthe eastern Alps. The number of simulated hail cases peaks in July and August,whereas large hail as reported in the ESWD is most prevalent between May and July. [ABSTRACT FROM AUTHOR]

Load forecasting is of crucial importance for operations of electric power systems. In recent years, deep learning based methods are emerging for load forecasting because their strong nonlinear approximation capabilities can provide more forecasting precision than conventional statistical methods. However, they usually suffer from some problems, e.g., the gradient vanishment and over-fitting. In order to address these problems, an unshared convolution based deep learning model with densely connected network is proposed. In this model, the backbone is the unshared convolutional neural network and a densely connected structure is adopted, which could alleviate the gradient vanishment. What is more, we use a regularization method named clipped $L_2$ -norm to overcome over-fitting, and design a trend decomposition strategy to address the possible distribution differences between the training and validation data. Finally, we conduct five case studies to verify the outperformance of our proposed deep learning model for deterministic and interval load forecasting. Two high-voltage and an medium-voltage real load datasets from Australia, Germany and America are used for model training and validation, respectively. Results show that the proposed model can achieve higher load forecasting accuracy, compared with other existing methods including the popular conventional methods such as naive forecast and generalized additive model, and deep learning methods, e.g., long short-term memory network, convolutional neural network, fully connected network, etc. [ABSTRACT FROM AUTHOR]