Classification Problem Research Articles

A pipeline for processing hyperspectral images, with a case of melanin-containing barley grains as an example.

Analysis of hyperspectral images is of great interest in plant studies. Nowadays, this analysis is used more and more widely, so the development of hyperspectral image processing methods is an urgent task. This paper presents a hyperspectral image processing pipeline that includes: preprocessing, basic statistical analysis, visualization of a multichannel hyperspectral image, and solving classification and clustering problems using machine learning methods. The current version of the package implements the following methods: construction of a confidence interval of an arbitrary level for the difference of sample averages; verification of the similarity of intensity distributions of spectral lines for two sets of hyperspectral images on the basis of the Mann-Whitney U-criterion and Pearson's criterion of agreement; visualization in two-dimensional space using dimensionality reduction methods PCA, ISOMAP and UMAP; classification using linear or ridge regression, random forest and catboost; clustering of samples using the EM-algorithm. The software pipeline is implemented in Python using the Pandas, NumPy, OpenCV, SciPy, Sklearn, Umap, CatBoost and Plotly libraries. The source code is available at: https://github.com/igor2704/Hyperspectral_images. The pipeline was applied to identify melanin pigment in the shell of barley grains based on hyperspectral data. Visualization based on PCA, UMAP and ISOMAP methods, as well as the use of clustering algorithms, showed that a linear separation of grain samples with and without pigmentation could be performed with high accuracy based on hyperspectral data. The analysis revealed statistically significant differences in the distribution of median intensities for samples of images of grains with and without pigmentation. Thus, it was demonstrated that hyperspectral images can be used to determine the presence or absence of melanin in barley grains with great accuracy. The flexible and convenient tool created in this work will significantly increase the efficiency of hyperspectral image analysis.

Methods and algorithms of swarm intelligence for the problems of nonlinear regression analysis and optimization of complex processes, objects, and systems: review and modification of methods and algorithms

The development of high-speed methods and algorithms for global multidimensional optimization and their modifications in various fields of science, technology, and economics is an urgent problem that involves reducing computing costs, accelerating, and effectively searching for solutions to such problems. Since most serious problems involve the search for tens, hundreds, or thousands of optimal parameters of mathematical models, the search space for these parameters grows non-linearly. Currently, there are many modern methods and algorithms of swarm intelligence that solve today's scientific and applied problems, but they require modifications due to the large spaces of searching for optimal model parameters. Modern swarm intelligence has significant potential for application in the energy industry due to its ability to optimize and solve complex problems. It can be used to solve scientific and applied problems of optimizing energy consumption in buildings, industrial complexes, and urban systems, reducing energy losses, and increasing the efficiency of resource use, as well as for the construction of various elements of energy systems in general. Well-known methods and algorithms of swarm intelligence are also actively applied to forecast energy production from renewable sources, such as solar and wind energy. This allows better management of energy sources and planning of their use. The relevance of modifications of methods and algorithms is due to the issues of speeding up their work when solving machine learning problems, in particular, in nonlinear regression models, classification, and clustering problems, where the number of observed data can reach tens and hundreds of thousands or more. The work considers and modifies well-known effective methods and algorithms of swarm intelligence (particle swarm optimization algorithm, bee optimization algorithm, differential evolution method) for finding solutions to multidimensional extremal problems with and without restrictions, as well as problems of nonlinear regression analysis. The obtained modifications of the well-known classic effective methods and algorithms of swarm intelligence, which are present in the work, effectively solve complex scientific and applied tasks of designing complex objects and systems. A comparative analysis of methods and algorithms will be conducted in the next study on this topic. Keywords: optimization, swarm intelligence, mathematical modelling, nonlinear regression, complex objects and systems.

Perceptron Theory Can Predict the Accuracy of Neural Networks.

Multilayer neural networks set the current state of the art for many technical classification problems. But, these networks are still, essentially, black boxes in terms of analyzing them and predicting their performance. Here, we develop a statistical theory for the one-layer perceptron and show that it can predict performances of a surprisingly large variety of neural networks with different architectures. A general theory of classification with perceptrons is developed by generalizing an existing theory for analyzing reservoir computing models and connectionist models for symbolic reasoning known as vector symbolic architectures. Our statistical theory offers three formulas leveraging the signal statistics with increasing detail. The formulas are analytically intractable, but can be evaluated numerically. The description level that captures maximum details requires stochastic sampling methods. Depending on the network model, the simpler formulas already yield high prediction accuracy. The quality of the theory predictions is assessed in three experimental settings, a memorization task for echo state networks (ESNs) from reservoir computing literature, a collection of classification datasets for shallow randomly connected networks, and the ImageNet dataset for deep convolutional neural networks. We find that the second description level of the perceptron theory can predict the performance of types of ESNs, which could not be described previously. Furthermore, the theory can predict deep multilayer neural networks by being applied to their output layer. While other methods for prediction of neural networks performance commonly require to train an estimator model, the proposed theory requires only the first two moments of the distribution of the postsynaptic sums in the output neurons. Moreover, the perceptron theory compares favorably to other methods that do not rely on training an estimator model.

Enhancing Skin Disease Diagnosis Through Deep Learning: A Comprehensive Study on Dermoscopic Image Preprocessing and Classification

ABSTRACTSkin cancer occurs when abnormal cells in the top layer of the skin, known as the epidermis, undergo uncontrolled growth due to unrepaired DNA damage, leading to the development of mutations. These mutations lead to rapid cell growth and development of cancerous tumors. The type of cancerous tumor depends on the cells of origin. Overexposure to ultraviolet rays from the sun, tanning beds, or sunlamps is a primary factor in the occurrence of skin cancer. Since skin cancer is one of the most common types of cancer and has a high mortality, early diagnosis is extremely important. The dermatology literature has many studies of computer‐aided diagnosis for early and highly accurate skin cancer detection. In this study, the classification of skin cancer was provided by Regnet x006, EfficientNetv2 B0, and InceptionResnetv2 deep learning methods. To increase the classification performance, hairs and black pixels in the corners due to the nature of dermoscopic images, which could create noise for deep learning, were eliminated in the preprocessing step. Preprocessing was done by hair removal, cropping, segmentation, and applying a median filter to dermoscopic images. To measure the performance of the proposed preprocessing technique, the results were obtained with both raw images and preprocessed images. The model developed to provide a solution to the classification problem is based on deep learning architectures. In the four experiments carried out within the scope of the study, classification was made for the eight classes in the dataset, squamous cell carcinoma and basal cell carcinoma classification, benign keratosis and actinic keratosis classification, and finally benign and malignant disease classification. According to the results obtained, the best accuracy values of the experiments were obtained as 0.858, 0.929, 0.917, and 0.906, respectively. The study underscores the significance of early and accurate diagnosis in addressing skin cancer, a prevalent and potentially fatal condition. The primary aim of the preprocessing procedures was to attain enhanced performance results by concentrating solely on the area spanning the lesion instead of analyzing the complete image. Combining the suggested preprocessing strategy with deep learning techniques shows potential for enhancing skin cancer diagnosis, particularly in terms of sensitivity and specificity.

Incremental minimum distortion embedding for online structural damage classification.

The damage detection problem in structures diminishes the maintenance cost and extends the life of the structures. Online structural damage classification is a relevant topic in structural health monitoring since monitoring the structure using sensors can help decision-makers can rely on empirical evidence for making informed choices regarding the maintenance, repair, or replacement of structural components. Sensors generate high-dimensional data. To address this challenge, employing dimensionality reduction methods becomes essential. These methods help obtain a low-dimensional representation of the data. Traditional dimensionality reduction methods are trained on a fixed data set, and incorporating new data may require retraining the entire model. This study employed the Minimum Distortion Embedding (MDE) method incrementally, ensuring the maintenance of a consistent embedding as new data becomes available. This incremental embedding took advantage of an anchor constraint to pin the existing embedding in place, then embed the new points. An online structural classification methodology was developed using the incremental MDE and a stream based hoeffding tree classifier. Stream data processing differs from batch processing, as it involves the real-time consumption of each incoming data. The developed methodology was tested with data obtained by accelerometers in a laboratory scaled jacket-type wind turbine foundation. The structural damage classification problems had 5 different classes including the undamaged class and 4 different damage classes. The damage corresponded to a 5mm crack in four different structural elements of the wind turbine foundation. The results indicate the good behavior of the damage classification methodology supported by the high values of the classification metrics obtained.

Testing the consistency of performance scores reported for binary classification problems

Binary classification is a fundamental task in machine learning, with applications across various scientific domains. Whether conducting fundamental research or refining practical applications, scientists typically assess and rank classification techniques based on performance metrics such as accuracy, sensitivity, and specificity. However, reported performance scores may not always provide a reliable basis for research ranking. The unreliability can be attributed to undisclosed or unconventional practices related to cross-validation, typographical errors, and other factors.In a given experimental setup with a specific number of positive and negative test items, performance scores can only assume specific, interrelated values. Based on this observation, in this paper, we introduce numerical techniques to assess the consistency of reported performance scores with the assumed experimental setup. Importantly, the proposed approaches do not rely on statistical inference. Instead, they use numerical methods (interval computing and integer linear programming) to identify inconsistencies with certainty.Through three applications in different fields of medicine, we demonstrate how the proposed tests can detect inconsistencies, thereby safeguarding the integrity of research fields. The power analyses of the tests in these applications show at least 71% of power when the performance scores are reported to four decimal places. In the investigated areas, the tests have so far identified inconsistencies in more than 100 scientific papers. To benefit the scientific community, we have made the consistency tests available in the open-source Python package mlscorecheck.

Quantitative characterization of fatigue damage in plate structures based on FSOM

For the problem of fatigue damage detection and damage degree assessment of plate structures, a quantitative damage assessment method based on the fast self-organizing feature mapping (FSOM) algorithm is proposed in this paper. The damage detection problem is transformed into a binary classification problem by extracting multidimensional damage features of the Lamb wave signal in plate to be detected and selecting damage sensitive features. Then, the FSOM network is used to identify the health state of the plate to be inspected, and the damage index is obtained by fusing the damage sensitive features using FSOM to quantitatively evaluate the damage level of the plate to be inspected. Simulation and experimental results show this method has a good dynamic tracking capability for the fatigue damage evolution of aluminum and composite plates, and can achieve quantitative assessment of fatigue damage of plate structures.

An improved long short-term memory with denoising autoencoder for solving text classification problems

In the past decade, text classification has been a popular research field for automatically categorizing documents into relevant categories. Predicting classes for input samples based on their features is the goal of classification. To limit the dimension of the feature space for precise text categorization, feature selection approaches are widely used for recognizing important and affective features while ignoring unimportant, weak, and chaotic components. This paper proposes a new model referred to as DAE-LSTM for reducing data dimension through the use of a denoising autoencoder. (Long Short-Term Memory (LSTM) is one of the popular text classification approaches, particularly when dealing with sequential datasets. Additionally, the Rectified Linear Unit (Relu) activation function was used instead of the hyperbolic tangent activation function (tanh) to improve the accuracy of the model. Four benchmark datasets were used to evaluate the proposed model against 4 cutting-edge text classification methods, including conventional Bidirectional LSTM, Bidirectional GRU, CNN-LSTM, and CNN-GRU. The suggested model has been found to perform much better than current methods using several kinds of performance assessment measures.

An automated ensemble approach using Harris Hawk optimization for visually evoked EEG signal classification.

Steady-state visually evoked potential is one of the active explorations in the brain-computer interface research. Electroencephalogram based brain computer interface studies have been widely applied to perceive solutions for real-world problems in the healthcare domain. The classification of externally bestowed visual stimuli of different frequencies on a human was experimented to identify the need of paralytic people. Although many classifiers are at the fingertip of machine learning technology, recent research has proven that ensemble learning is more efficacious than individual classifiers. Despite its efficiency, ensemble learning technology exhibits certain drawbacks like taking more time on selecting the optimal classifier subset. This research article utilizes the Harris Hawk Optimization algorithm to select the best classifier subset from the given set of classifiers. The objective of the research is to develop an efficient multi-classifier model for electroencephalogram signal classification. The proposed model utilizes the Boruta Feature Selection algorithm to select the prominent features for classification. Thus selected prominent features are fed into the multi-classifier subset which has been generated by the Harris Hawk Optimization algorithm. The results of the multi-classifier ensemble model are aggregated using Stacking, Bagging, Boosting, and Voting. The proposed model is evaluated against the acquired dataset and produces a promising accuracy of 96.1%, 98.7%, 91.91%, and 99.01% with the ensemble techniques respectively. The proposed model is also validated with other performance metrics such as sensitivity, specificity, and F1-Score. The experimental results show that the proposed model proves its supremacy in segregating the multi-class classification problem with high accuracy.

MACHINE LEARNING METHODS FOR BREAST CANCER CLASSIFICATION BY USING DATA SCIENCE TECHNIQUES

This study explores the sensitivity analysis of various machine learning methods applied to the problem of breast cancer classification. By examining the robustness and performance of different algorithms, we aim to identify the most reliable techniques for accurate diagnosis. We assess the impact of key parameters and data variations on model outcomes to provide a comprehensive understanding of each method's strengths and limitations. Our findings offer valuable insights into the selection and optimization of machine learning models for breast cancer detection, ultimately contributing to improved diagnostic accuracy and patient care.

A hybrid approach to obesity level determination with decision tree and pelican optimization algorithm

Approximately 2 billion people in the world struggle with "obesity" and factors like eating lifestyle, habits, health conditions and mode of transport affect obesity. In this study, an artificial intelligence and machine learning-based model has been developed to predict obesity levels. It is proposed to create a hybrid model by combining the Decision Tree (DT) algorithm with the Pelican Optimization Algorithm (POA) on the obesity dataset of 2111 patients in SSggle. These models emphasize the critical role of parameters, aiming to achieve high performance. To solve the classification problem of multi-class obesity level determination, fuzzy logic-based parameter optimization is used to achieve high performance. While obesity rates are increasing worldwide, the study, which aims to globalize the parameters with the random discovery strategy of POA, is thought to be helpful for health professionals and decision-makers by successfully predicting obesity levels.

Anomaly-based error and intrusion detection in tabular data: No DNN outperforms tree-based classifiers

Recent years have seen a growing involvement of researchers and practitioners in crafting Deep Neural Networks (DNNs) that seem to outperform existing machine learning approaches for solving classification problems as anomaly-based error and intrusion detection. Undoubtedly, classifiers may be very diverse among themselves, and choosing one or another is typically due to the specific task and target system. Designing and training the optimal tabular data classifier requires extensive experimentation, sensitivity analyses, big datasets, and domain-specific knowledge that may not be available at will or considered a non-strategical asset by many companies and stakeholders. This paper compares, using a total of 23 public datasets: i) traditional (tree-based, statistical) supervised classifiers, ii) DNNs that are specifically designed for classifying tabular data, iii) DNNs for image classification that are applied to tabular data after converting data points into images, alone and as ensembles. Experimental results and related discussions show clear advantages in adopting tree-based classifiers for anomaly-based error and intrusion detection in tabular data as they outperform their competitors, including DNNs. Then, individual classifiers are compared against ensembles using different combinations of the classifiers considered in this study as base-learners, providing a unified final response through many meta-learning strategies. Results show that there is no benefit in building ensembles instead of using a tree-based classifier as Random Forests, eXtreme Gradient Boosting or Extra Trees. The paper concludes that anomaly-based error and intrusion detectors for critical systems should use the old (but gold) tree-based classifiers, which are also easier to fine-tune, and understand; plus, they require less time and resources to learn their model.

SSAL-Net: Semi-supervised network integrating self-supervised adversarial learning for diagnosing subtypes of pulmonary nodules

Accurately diagnosing the subtype of pulmonary nodules is a crucial step in screening patients with early-stage lung cancer. However, most of existing researches can only diagnose benign and malignant pulmonary nodules but cannot give detailed subtypes. Besides, although convolutional neural networks (CNNs) have been widely used in the intelligent diagnosis of benign and malignant pulmonary nodules, it is difficult to further improve the diagnostic accuracy due to the small number of samples. In this paper, a semi-supervised network integrating self-supervised adversarial learning (SSAL-Net) suitable for small sample classification problem is proposed to diagnose adenocarcinoma, squamous cell carcinoma, inflammation, and other benign types of pulmonary nodules. SSAL-Net is mainly composed of a self-supervised classification network (SSC-Net), an unsupervised adversarial reconstruction network (UAR-Net) and a supervised classification network (SC-Net), where SSC-Net and UAR-Net share an encoder to provide representative image features to SC-Net. SSAL-Net takes three different forms of pulmonary nodule data generated based on prior knowledge as input, and enables end-to-end collaborative training through SSC-Net, UAR-Net and SC-Net performing their respective tasks. The experimental results show that SSAL-Net can identify adenocarcinoma, squamous cell carcinoma, inflammation and other benign types of pulmonary nodules with an average accuracy of 89.29% on our self-constructed lung CT image dataset, and can identify malignant pulmonary nodules with an average accuracy of 94.67% on the publicly available LIDC-IDRI dataset, obtaining the state-of-the-art results.

Разработка и исследование методов классификации территории по интенсивности выпадения осадков

The article describes the implementation of the weather classification service in the territory and the use of a deep neural network to solve the classification problem. It is noted that deep neural networks have a high efficiency index with a fairly complete set of training data. The deep neural network model is based on data that was received from the ERA-5 satellite, the data consists of 11 parameters such as latitude, longitude, time, wind components, temperature, pressure and others. The analysis of the existing methods of weather classification is also carried out. The article describes the application of a deep neural network in the context of weather forecasting, and also includes consideration of the importance of having a sufficient amount of training data for the successful application of machine learning methods

Алгоритм аппроксимации сигналов для классификации электроэнцефалограмм человека

The article examines the problem of classification of electroencephalograms (EEG), where noise in the signals, caused by various factors, prevents effective analysis and interpretation of the data. The main goal of the study is to analyze the effectiveness of a signal approximation algorithm using a wavelet technique with the next piece-wise approximation in order to effectively remove noise and subsequently solve the problem of signal classification using a convolutional neural network. The classification accuracy of the proposed algorithm with a low-pass filter is compared at different cutoff frequencies. One of the key findings of the study is a two-step approach to signal processing. In the first stage, the model is trained on the raw data, and then the trained convolution kernel with the highest variance is applied to the original signals. This solves the problem of choosing the mother function. This approach aims to enhance the informative components in the signal. The second processing step involves applying the proposed approximation algorithm after the convolution, which complements the first step to create a comprehensive method. This method not only effectively reduces noise in the data, but also has high potential to improve the overall accuracy in solving EEG signal classification problems. Thus, the results of the study provide important practical and theoretical implications, highlighting the prospects for applying the proposed method in the field of signal analysis.

The complete chloroplast genome of the Syzygium buxifolium Hook. Et Arn.1833 and its phylogenetic analysis

Syzygium buxifolium. Hook. Et Arn.1833 is a member of the Myrtaceae family. This species is used in traditional Chinese medicines. It possesses numerous synonyms, reflecting the ambiguity in its taxonomy. The chloroplast genome has been widely used for species identification and phylogenetic analysis. Regrettably, there is a lack of information regarding the chloroplast genome of S. buxifolium. Here, we intend to obtain the chloroplast genome of S. buxifolium to resolve its classification problems. In particular, we utilized Illumina sequencing technology to sequence, GetOrganelle to assemble, and CPGAVAS2 to characterize the chloroplast genome of S. buxifolium. The chloroplast genome of S. buxifolium had a length of 158,581 bp and consisted of 111 unique genes, comprising 78 protein-coding genes, 29 transfer RNA (tRNA) genes, and four ribosomal RNA (rRNA) genes. In addition, we identified 86 Simple Sequence Repeats, 345 tandem repetitive sequences, and 34 dispersed repetitive sequences using modules implemented in CPGAVAS2. Lastly, we carried out phylogenetic analysis using Phylosuite. The results indicated a close relationship between S. buxifolium and S. grijsii. This study offers novel genetic data for the molecular identification and subsequent phylogenetic analysis of the Syzygium genus.

An integrated approach for detecting and classifying pores and surface topology for fatigue assessment 316L manufactured by powder bed fusion of metals using a laser beam using μCT and machine learning algorithms

AbstractThis research aims to detect and analyze critical internal and surface defects in metal components manufactured by powder bed fusion of metals using a laser beam (PBF‐LB/M). The aim is to assess their impact on the fatigue behavior. Therefore, a combination of methods, including image processing of micro‐computed tomography ( CT) scans, fatigue testing, and machine learning, was applied. A workflow was established to contribute to the nondestructive assessment of component quality and mechanical properties. Additionally, this study illustrates the application of machine learning to address a classification problem, specifically the categorization of pores into gas pores and lack of fusion pores. Although it was shown that internal defects exhibited a reduced impact on fatigue behavior compared with surface defects, it was noted that surface defects exert a higher influence on fatigue behavior. A machine learning algorithm was developed to predict the fatigue life using surface defect features as input parameters.

Manifold neighboring envelope sample generation mechanism for imbalanced ensemble classification

For existing imbalanced ensemble (IE) methods, the sample subsets are constructed from the same dataset, which usually suffer from low quality (diversity and separability) of the subsets, so a manifold neighboring envelope sample generation mechanism (MNESG) and an imbalanced ensemble algorithm based on the mechanism (MNESG-IE) are proposed to solve this problem. First, for the original balanced subsets (OBS), a manifold neighboring sample envelope projection mechanism (MNSEP) is designed to mine the local correlation information between the samples and their nearest neighbors in the subsets. Second, the fuzzy c-means clustering (FCM) is used to further mine the global correlation information among similar samples in the subsets. Third, the sample distribution consistency preservation mechanism (SDCPM) is designed to enhance the consistency of the sample distribution before and after clustering. To better reduce the three accumulated losses above, the three steps are conducted simultaneously, thereby realizing the MNESG, which can transform the OBS into two new types of high quality envelope sample subsets − neighboring envelope sample (NES) subsets and neighboring cluster envelope sample (NCES) subsets. Finally, base classifiers are trained on the NES subsets and NCES subsets, and then fused by a two-dimensional sparse fusion mechanism (2D-SFM). Various representative IE algorithms on over thirty benchmark datasets are considered for verification. The results show that compared with the state-of-the-art IE algorithms, MNESG-IE achieves 17.79%, 17.90%, 23.61%, 18.08% improvement in terms of ACC, AUC, F-M and G-M, respectively. The major originality of the paper is: (a) proposing the MNSEP to mine the local correlation information for improving the quality of the subsets; (b) proposing the MNESG to generate high quality subsets by mining local and global correlation information simultaneously; and (c)forming an IE algorithm to better solve the imbalanced classification problem.

Using SAR Data for Monitoring of Agricultural Crops in the South of the Russian Far East

The use of SAR data to monitoring agricultural crops is a promising area of research designed to complement existing methods and technologies based on the analysis of multispectral images. The main advantages of vegetation indices calculated from SAR data include their sensitivity to the polarimetric properties of the backscatter intensity, its scattering characteristics, and independence from cloud cover. This is especially important for the territory of the south of the Russian Far East, whose monsoon climate provides humid and cloudy weather during the period when crops gain maximum biomass. For arable lands in the Khabarovsk Territory and the Amur Region, a total of 64 Sentinel-1 SAR images were obtained from May to October 2021. For each date, the values of the DpRVI, RVI, VH/VV indices were calculated and time series were constructed for the entire observation period for individual fields (342 fields in total). NDVI time series were constructed from Sentinel-2 multispectral images using a cloud mask. The characteristics of time series extremes were calculated for different types of arable land: soybeans, oats, and fallows. It was shown that for each crop the seasonal curves DpRVI, RVI, VH/VV had a characteristic appearance. It was found that the DpRVI demonstrated the highest stability – the coefficients of variation of the seasonal variation of DpRVI were significantly lower than those for RVI and VH/VV. It was also revealed that the similarity between the curves of these indices remained for regions quite distant from each other - the Khabarovsk Territory and the Amur Region. The main characteristics of the seasonal variation of time series of radar indices were calculated in comparison with NDVI - the magnitude of the maximum, the date of the maximum and the values of the coefficient of variation for these indicators. It was found, firstly, that the values of these indicators in different regions are similar to each other; secondly, the variability of the maximum and the day of the maximum for DpRVI is lower than for RVI and VH/VV; thirdly, the variability of the maximum and the day of the maximum for DpRVI is comparable to NDVI. Thus, time series of radar indices DpRVI, RVI, VH/VV for the main types of agricultural lands in the south of the Far East have distinctive features and can be used in classification problems, yield modeling and crop rotation control.

Unet-boosted Classifier – Multi-Task Architecture for Small Datasets Applied to Brain MRI Classification

The problem of training deep neural networks on small samples is especially relevant for medical issues. The paper examines the impact of pixel-wise marking of significant objects in the image, over the true class label, on the quality of the classification. To achieve better classification results on small samples, we propose a multitasking architecture – Unet-boosted classifier (UBC), that is trained simultaneously to solve classification and semantic segmentation problems. As the exploratory dataset, MRI images of patients with benign glioma and glioblastoma taken from the BRaTS 2019 data set are used. One horizontal slice of the MRI image containing a glioma is considered as the input (a total of 380 frames in the training set), and the probability of glioblastoma – as the output. Resnet34 was chosen as the baseline, trained without augmentations with a loss function based on cross-entropy. As an alternative solution, UBC-resnet34 is used – the same resnet34, boosted by a decoder built on the U-Net principle and predicting the pixels with glioma. The smoothed Sorensen-Dice coefficient (DiceLoss) is used as a decoder loss function. Results on the test sample: accuracy for the baseline reached 0.71, for the proposed model – 0.81, and the Dice score – 0.77. Thus, a deep model can be well trained even on a small data set, using the proposed architecture, provided that marking of the affected tissues in the form of a semantic mask is provided.