The paper proposes a technology for effective feature selection to localize individual characteristics of anatomical and pathological structures in the human eye fundus. Such an approach allows the intellectual analysis of features to be conducted using color subspaces and the regions of interest to be identified. This problem is relevant because in this way the efficiency of laser coagulation surgery can be improved. The technology is based on the texture analysis of certain image patterns. The initial textural attributes are derived from different statistical image descriptors calculated using the MaZda library (image histogram, image gradient, series length and adjacency matrices). The analysis of the feature space informativity and selection of the most effective features are carried out using the discriminant data analysis. The best-size image fragmentation windows for eye fundus clustering and sets of features that provide the necessary accuracy in identifying the regions of interest were derived via analyzing the following four image classes: exudates, thick vessels, thin vessels, and healthy areas. The feature selection technology was based on clustering using a K-means method, with the Euclidean and Mahalanobis distance used as a similarity measure. The required minimum size of the fragmentation window and the similarity measure were chosen from a criterion of the minimum clustering error among all the smallest window sizes. The article also presents a system for automatically forming a coagulate plan, expected to be used to support the decision-making during laser retinal coagulation surgery in the treatment of diabetic macular edema. This system is currently being developed based on the proposed technology.
The study was devoted to the analysis of wavefront aberrations under changes of the cornea surface curvature in the human eye. The analysis was based on the representation of aberrations of the front and rear corneal surfaces as superposition of Zernike functions. Weight coefficients of the Zernike polynomials were the object of this study. The data under analysis were obtained in a number of clinical trials in the Branchevski’s eуе clinic. The most informative weight coefficients were analyzed from the point of view of classification of patients by particular diagnosis. A comparison of the classification results was carried out using thirty front and rear corneal features, as well as the most informative features. The features were ranked by the informativity criterion for solving a specific classification task. While doing analysis, the informativity was evaluated on the basis of values of a separability criterion. An additional estimation of informativeness was carried out by calculating the classification error by a K-means method. As a result of the analysis, basic Zernike functions that are most informative for particular eye pathologies were identified.
Detailed automated analysis of crop images is critical to the development of smart agriculture and can significantly improve the quantity and quality of agricultural products. A hyperspectral camera potentially allows to extract more information about the observed object than a conventional one, so its use can help in solving problems that are difficult to solve with conventional methods. Often, predictive models that solve such problems require a large dataset for training. However, sufficiently large datasets of hyperspectral images of agricultural plants are not currently publicly available. Therefore, we present a new dataset of hyperspectral images of plants in this paper. This dataset can be accessed via URL https://pypi.org/project/HSI-Dataset-API/. It contains 385 hyperspectral images with a spatial resolution of 512 by 512 pixels and spectral resolution of 237 spectral bands. The images were captured in the summer of 2021 in Samara and Novocherkassk (Russia) using Offner based Imaging Hyperspectrometer of our own production. The article demonstrates the work of some basic approaches to the analysis of hyperspectral images using the dataset and states problems for further solving.
Advances in the neural networks have brought revolution in many areas, especially those related to image processing and analysis. The most complex is a task of analyzing biomedical data due to a limited number of samples, imbalanced classes, and low-quality labelling. In this paper, we look into the possibility of using neural networks when solving a task of semantic segmentation of fundus. The applicability of the neural networks is evaluated through a comparison of image segmentation results with those obtained using textural features. The neural networks are found to be more accurate than the textural features both in terms of precision (~25%) and recall (~50%). Neural networks can be applied in biomedical image segmentation in combination with data balancing algorithms and data augmentation techniques.