Search

This paper presents a novel method for improving the invariance of convolutional neural networks (CNNs) to selected geometric transformations in order to obtain more efficient image classifiers. A common strategy employed to achieve this aim is to train the network using data augmentation. Such a method alone, however, increases the complexity of the neural network model, as any change in the rotation or size of the input image results in the activation of different CNN feature maps. This problem can be resolved by the proposed novel convolutional neural network models with geometric transformations embedded into the network architecture. The evaluation of the proposed CNN model is performed on the image classification task with the use of diverse representative data sets. The CNN models with embedded geometric transformations are compared to those without the transformations, using different data augmentation setups. As the compared approaches use the same amount of memory to store the parameters, the improved classification score means that the proposed architecture is more optimal.

Image analysis has many practical applications and proper representation of image content is its crucial element. In this work, a novel type of representation is proposed where an image is reduced to a set of highly sparse matrices. Equivalently, it can be viewed as a set of local features of different types, as precise coordinates of detected keypoints are given. Additionally, every keypoint has a value expressing feature intensity at a given location. These features are extracted from a dedicated convolutional neural network autoencoder. This kind of representation has many advantages. First of all, local features are not manually designed but are automatically trained for a given class of images. Second, as they are trained in a network that restores its input on the output, they may be expected to minimize information loss. Consequently, they can be used to solve similar tasks replacing original images, such an ability was illustrated with image classification task. Third, the generated features, although automatically synthesized, are relatively easy to interpret. Taking a decoder part of our network, one can easily generate a visual building block connected with a specific feature. As the proposed method is entirely new, a detailed analysis of its properties for a relatively simple data set was conducted and is described in this work. Moreover, to present the quality of trained features, it is compared with results of convolutional neural networks having a similar working principle (sparse coding).

The QSAR models for a set of pyrazolo[4,3-e][1,2,4]triazines incorporating benzenesulfonamide moiety combined directly with the heterocyclic ring or by NH linkage were generated. The inhibitory potency of compounds against human carbonic anhydrase isoforms IX and XII and antiproliferative activity against human MCF-7 cells were used as the dependent variables. The Codessa pro software was used for the descriptors calculation and the Best Multi-Linear Regression (BMLR) algorithm was employed to build the QSAR models. It was found that quantum descriptors are critical of the compounds activities. The selected models have good predictive accuracy confirmed by a set of the statistical quantities recommended by OECD.

A new series of sulfonamide derivatives of pyrazolo[4,3-e][1,2,4]triazine with chiral amino group has been synthesized and characterized. The compounds were tested for their tyrosinase and urease inhibitory activity. Evaluation of prepared derivatives demonstrated that compounds (8b) and (8j) are most potent mushroom tyrosinase inhibitors whereas all of the obtained compounds showed higher urease inhibitory activity than the standard thiourea. The compounds (8a), (8f) and (8i) exhibited excellent enzyme inhibitory activity with IC50 0.037, 0.044 and 0.042 μM, respectively, while IC50 of thiourea is 20.9 μM.

Background Triazoles and their fused derivatives are an important class of compounds that exhibit interesting biological properties, such as antiasthmatic, antimicrobial, antifungal, analgesic, antiallergic, antiinflammatory, herbicidal, plant growth regulative activity, and anti-HIV-1 activities. Moreover, anticancer activity of 1,2,4-triazole containing derivatives has been documented. Due to the fact a convenient approach toward polycyclic frameworks containing fused 1,2,4-triazoles was described. Objective The objective of this article is the synthesis of new pyrazolo[4,3-e]triazolo[4,5- b][1,2,4]triazine derivatives with potential antiproliferative activity. Methods Cancer cell proliferation was analysed by means of MTT assay after 96 h treatment. IC50 was calculated using computerized linear regression analysis of quantal log doseprobit functions, according to the method of Litchfield and Wilcoxon. X-ray data were collected on the Bruker SMART APEX II CCD diffractometer; The structure was solved by direct methods using SHELXS-2013 and refined by full-matrix least-squares with SHELXL-2014/7. All calculations were performed using WINGX version 2014.1 package. Results The series of pyrazolo[4,3-e]triazolo[4,5-b][1,2,4]triazine derivatives were synthesized. MTT assay revealed that the compounds inhibited cancer cells growth at concentrations below 10 µM. The tested compounds showed higher antiproliferative activity than popular cytostatics cisplatin (lung carcinoma) and 5-fluorouracil (colon adenocarcinoma). X-ray examinations showed that final products in the crystalline phase have a linear form. Conclusion In the paper we have reported the synthesis and spectroscopic analysis of new condensed tricyclic derivatives of the pyrazolo[4,3-e]triazolo[4,5-b][1,2,4]triazine. MTT analysis revealed concentration-dependent decrease in lung A549 and colon LS180 cancer cells proliferation. In order to explain the molecular mechanisms involved in anticancer activity of pyrazolo[4,3- e]triazolo[4,5-b][1,2,4]triazine derivatives, our research will be continued.

In this paper a new architecture of convolutional neural networks is proposed. It is a fully-convolutional architecture which allows to keep the size of the processed image constant. This, in consequence, allows to apply it for image segmentation tasks where for a given image a mask representing sought regions should be produced. An additional advantage of this architecture is its ability to learn from smaller images which reduces the amount of data that must be propagated through the network. The trained network can be still applied to images of any size. The proposed method was used for automatic localization of demyelinating plaques in head MRI sequences. This work was possible, which should be emphasized, only thanks to the manually outlined plaques provided by radiologist. To present characteristic of the considered approach three architectures and three result evaluation methods were discussed and compared.