A Semiautomated Deep Learning Approach for Pancreas Segmentation

Accurate pancreas segmentation from 3D CT volumes is important for pancreas diseases therapy. It is challenging to accurately delineate the pancreas due to the poor intensity contrast and intrinsic large variations in volume, shape, and location. In this paper, we propose a semiautomated deformable U-Net, i.e., DUNet for the pancreas segmentation. The key innovation of our proposed method is a deformable convolution module, which adaptively adds learned offsets to each sampling position of 2D convolutional kernel to enhance feature representation. Combining deformable convolution module with U-Net enables our DUNet to flexibly capture pancreatic features and improve the geometric modeling capability of U-Net. Moreover, a nonlinear Dice-based loss function is designed to tackle the class-imbalanced problem in the pancreas segmentation. Experimental results show that our proposed method outperforms all comparison methods on the same NIH dataset.

Photonic resonator interferometric scattering microscopy

Interferometric scattering microscopy is increasingly employed in biomedical research owing to its extraordinary capability of detecting nano-objects individually through their intrinsic elastic scattering. To significantly improve the signal-to-noise ratio without increasing illumination intensity, we developed photonic resonator interferometric scattering microscopy (PRISM) in which a dielectric photonic crystal (PC) resonator is utilized as the sample substrate. The scattered light is amplified by the PC through resonant near-field enhancement, which then interferes with the <1% transmitted light to create a large intensity contrast. Importantly, the scattered photons assume the wavevectors delineated by PC’s photonic band structure, resulting in the ability to utilize a non-immersion objective without significant loss at illumination density as low as 25 W cm−2. An analytical model of the scattering process is discussed, followed by demonstration of virus and protein detection. The results showcase the promise of nanophotonic surfaces in the development of resonance-enhanced interferometric microscopies.

The learning-based automatic segmentation algorithm of brain MR images based on 7T

Background: The learning-based algorithms provide an ability to automatically estimate and refine GM, WM and CSF. The ground truth manually achieved from the 3T MR image may not be accurate and reliable with poor image intensity contrast. It will seriously influence the classification performance because the supervised learning-based algorithms extremely rely on the ground truth. Recently, the 7T MR images brings about the excellent image intensity contrast, while Structured Random Forest (SRF) performs the pixel-level classification and achieves structural and contextual information in images. Materials and Methods: In this paper, a automatic segmentation algorithm is proposed based on ground truth achieved by the corresponding 7T subjects for segmenting the 3T&1.5T brain tissues using SRF classifiers. Through taking advantage of the 7T brain MR images, we can achieve the highly accuracy and reliable ground truth and then implement the training of SRF classifiers. Our proposed algorithm effectively integrates the T1-weighed images along with the probability maps to train the SRF classifiers for brain tissue segmentation. Results: Specifically, for the mean Dice ratio of all 10 subjects, the proposed method achieved 95.14%±0.9%, 90.17%±1.83%, and 81.96%± 4.32% for WM, GM, and CSF. With the experiment results, the proposed algorithm can achieve better performances than other automatic segmentation methods. Further experiments are performed on the 200 3T&1.5T brain MR images of ADNI dataset and our proposed method shows promised performances. Conclusions: The authors have developed and validated a novel fully automated method for 3T brain MR image segmentation.

ALGORITMA KNN UNTUK KLASIFIKASI KEMATANGAN BUAH APEL BERDASARKAN TEKSTUR

An apple is a fruit that is widely planted in mountainous or cold regions, for example in Malang. Apples are fruits that have many colors, there are green, yellow, and red colors. With a variety of colors that make consumers feel confused whether the apples to be eaten are sweet or sour. Because almost most consumers do not know the type of apple that will be purchased. Sometimes the type of apple that will be purchased is green, but because it does not know ripe or raw, it is wrong to choose. In order to help consumers in knowing the level of maturity of apple again, a system was made. With the aim to be able to classify the level of maturity of the apple again. So that the system created will display information whether the apple is more ripe or raw. The system will process the image of the apple again and take the GLCM texture features (intensity, contrast, energy, smoothness, entropy, skewness). And the process of determining fruit maturity using the KNN method. The system was built using the matlab tool, with 200 datasets, consisting of 130 training datasets, and 70 testing datasets. In applying the KNN algorithm to determine the maturity level of apples, the accuracy results are 51.4%. With output data that is not in accordance with the target number of 34 data and according to the target number of 36 data.

Semi-dynamic Control of FCM Initialization for Automatic Extraction of Inflamed Appendix from Ultrasonography

Background: Current naked-eye examination of the ultrasound images for inflamed appendix has limitations due to its intrinsic operator subjectivity problem. Objective: In this paper, we propose a fully automatic intelligent method for extracting inflamed appendix from ultrasound images. Accurate and automatic extraction of inflamed appendix from ultrasonography is a major decision making resource of the diagnosis and management of suspected appendicitis. Methods: The proposed method uses Fuzzy C-means learning algorithm in pixel clustering with semi-dynamic control of initializing the number of clusters based on the intensity contrast dispersion of the input image. Thirty percent of the prepared ultrasonography samples are classified into four different groups based on their intensity contrast distribution and then different number of clusters are assigned to the images in accordance with such groups in Fuzzy C-means learning process. Results: In the experiment, the proposed system successfully extracts the target without human intervention in 82 of 85 cases (96.47% accuracy). The proposed method also shows that it can cover the false negative cases occurred previously that used self-organizing map as the learning engine. Conclusion: Such high level reliable correct extraction of inflamed appendix encourages to use the automatic extraction software in the diagnosis procedure of suspected acute appendicitis.

Determining a reliably visible and inexpensive surface fiducial marker for use in MRI: a research study in a busy Australian Radiology Department

ObjectivesSingle-use commercial surface fiducial markers are used in clinical imaging for a variety of applications. The current study sought to find a new, reliably visible, easily sourced and inexpensive fiducial marker alternative for use with MRI.DesignFive commonly requested MRI sequences were determined (three-dimensional (3D) T1-weighted, T1 coronal, 3D T2-weighted, T2 fat suppressed, proton density), to examine the visibility of 18 items (including a commercial fiducial marker).SettingClinical 3T MRI scanner in an Australian Tertiary Hospital and an Australian University Biomedical Engineering research group.Interventions18 marker alternatives were scanned using five common MRI sequences. Images were reformatted to obtain both an image through the mid-height of each marker and a maximum intensity z-projection image over the volume of the marker. Variations in marker intensity were profiled across each visible marker and a visibility rating defined.Main outcome measuresOutcome measures were based on quantitative assessment of a clear intensity contrast ratio between the marker and the adjacent tissue and a qualitative assessment of visibility via a 3-point scale.ResultsThe fish oil capsule, vitamin D capsule, paint ball pellet, soy sauce sushi tube and commercial markers were typically visible to a high quality on all the imaging sequences and demonstrated a clear differential in intensity contrast against the adjacent tissue. Other common items, such as plasticine ‘play doh’ and a soft ‘Jelly baby’ sweet, were surprise candidates, demonstrating high-quality visibility and intensity contrast for the 3D T1-weighted sequence.ConclusionsDepending on the basis for referral and MRI sequence chosen, four alternative fiducial markers were determined to be inexpensive, easily sourced and consistently visible. Of these, the vitamin D capsule provided an excellent balance between availability, size, cost, usability and quality of the visualised marker for all the commonly used MRI sequences analysed.

Intensity contrast of solar network and faculae

Aims. We aim to gain insight into the effect of network and faculae on solar irradiance from their apparent intensity. Methods. Taking full-disc observations from the Solar Dynamics Observatory, we examined the intensity contrast of network and faculae in the continuum and core of the Fe I 6173 Å line and 1700 Å, including the variation with magnetic flux density, distance from disc centre, nearby magnetic fields, and time. Results. The brightness of network and faculae is believed to be suppressed by nearby magnetic fields from its effect on convection. We note that the degree of magnetically crowding of an area also affects the magnetic flux tube sizes and the depth at which magnetic concentrations are embedded in intergranular lanes, such that intensity contrast can be enhanced in magnetically crowded areas at certain flux densities and distances from disc centre. The difference in intensity contrast between the quiet-Sun network and active region faculae, noted by various studies, arises because active regions are more magnetically crowded and is not due to any fundamental physical differences between network and faculae. These results highlight that solar irradiance models need to include the effect of nearby magnetic fields on network and faculae brightness. We found evidence that suggests that departures from local thermal equilibrium (LTE) might have limited effect on intensity contrast. This could explain why solar irradiance models that are based on the intensity contrast of solar surface magnetic features calculated assuming LTE reproduce the observed spectral variability even where the LTE assumption breaks down. Certain models of solar irradiance employ chromospheric indices as direct indications of the effect of network and faculae on solar irradiance. Based on past studies of the Ca II K line and on the intensity contrast measurements derived here, we show that the fluctuations in chromospheric emission from network and faculae are a reasonable estimate of the emission fluctuations in the middle photosphere, but not of those in the lower photosphere. This is due to the different physical mechanisms that underlie the magnetic intensity enhancement in the various atmospheric regimes, and represents a fundamental limitation of these solar irradiance models. Any time variation in the radiant properties of network and faculae is, of course, relevant to their effect on solar irradiance. The data set, which extends from 2010 to 2018, indicates that their intensity contrast was stable to about 3% in this period. Conclusions. This study offers new insights into the radiant behaviour of network and faculae, with practical implications for solar irradiance modelling.

Intensity contrast of solar plage as a function of magnetic flux at high spatial resolution

Magnetic elements have an intensity contrast that depends on the type of region they are located in (for example quiet Sun, or active region plage). Observed values also depend on the spatial resolution of the data. Here we investigate the contrast-magnetic field dependence in active region plage observed near disk center with SUNRISE during its second flight in 2013. The wavelengths under study range from the visible at 525 nm to the near ultraviolet (NUV) at 300 nm and 397 nm. We use quasi-simultaneous spectropolarimetric and photometric data from the Imaging Magnetograph eXperiment (IMaX) and the Sunrise Filter Imager (SuFI), respectively. We find that in all wavelength bands, the contrast exhibits a qualitatively similar dependence on the line-of-sight magnetic field, BLOS, as found in the quiet Sun, with the exception of the continuum at 525 nm. There, the contrast of plage magnetic elements peaks for intermediate values of BLOS and decreases at higher field strengths. By comparison, the contrast of magnetic elements in the quiet Sun saturates at its maximum value at large BLOS. We find that the explanation of the turnover in contrast in terms of the effect of finite spatial resolution of the data is incorrect with the evidence provided by the high-spatial resolution SUNRISE data, as the plage magnetic elements are larger than the quiet Sun magnetic elements and are well-resolved. The turnover comes from the fact that the core pixels of these larger magnetic elements are darker than the quiet Sun. We find that plages reach lower contrast than the quiet Sun at disk center at wavelength bands formed deep in the photosphere, such as the visible continuum and the 300 nm band. This difference decreases with formation height and disappears in the Ca II H core, in agreement with empirical models of magnetic element atmospheres.