Research

Deep learning for medical diagnostics — biomedical image analysis, generative models for synthetic training data, and neural architecture search.

citations

1,804

h-index

8

i10-index

7

• Google Scholar
• Scopus
• ORCID
• ResearchGate

20257

• Devising a comprehensive approach to diagnosing breast cancer subtypes automatically based on deep neural networks

  Oleh Berezsky, Pavlo Liashchynskyi, Petro Liashchynskyi, Petro Selskyy

  Eastern-European Journal of Enterprise Technologies

  Tackles the subjectivity of scoring immunohistochemical biomarkers in breast cancer, where assessments vary between individual pathologists. Evolutionary algorithms tune deep networks to balance model complexity against generalisation, evaluated by segmenting immunohistochemical images across 13 different architectures.

  doi:10.15587/1729-4061.2025.344041

• Combined metric for evaluating the quality of synthesized biomedical images

  Oleh Berezsky, Mykhailo Berezkyi, Mykola Dombrovskyi, Petro Liashchynskyi, Grygoriy Melnyk

  Radio Electronics, Computer Science, Control

  Proposes a new combined metric for judging how good a synthesized image actually is — a prerequisite for using generated data in training. Also examines the potential of diffusion models for biomedical image synthesis.

  doi:10.15588/1607-3274-2025-2-15

• Software system for automatic diagnosis of breast cancer

  Oleh Berezsky, Petro Liashchynskyi

  Ukrainian Journal of Information Technology

  Design of an end-to-end software system for automatic breast cancer diagnosis, aimed at improving both the accuracy and the speed of oncological diagnosis.

  doi:10.23939/ujit2025.02.035

• Biomedical Image Datasets

  Oleh Berezsky, Grygoriy Melnyk, Petro Liashchynskyi, Oleh Pitsun

  Intellectual Systems of Decision Making (Springer, LNDECT)· 2 citations

  A book chapter surveying the biomedical image datasets available for training diagnostic models, and the characteristics that make them usable — or not — for deep learning.

  doi:10.1007/978-3-031-88483-2_3

• Integrating convolutional neural networks and autoencoders for skin lesion diagnosis

  Kateryna Bilyk, Olha Narushynska, Petro Liashchynskyi, Vasyl Teslyuk

  MoMLeT 2025 (CEUR Workshop Proceedings)

  Combines convolutional classifiers with autoencoders for the diagnosis of skin lesions, extending the group’s biomedical imaging work from histology into dermatology.

• An overview of deep learning methods for automatic reporting in digital pathology

  Anastasiia Fedenkiv, Petro Liashchynskyi

  CISS-2025 — Computational Intelligence and Smart Systems, Lviv Polytechnic

  A systematic review of image-to-text methods that generate reports automatically from histopathological slides. Where earlier reviews stay descriptive, this one targets the interpretability gap between available tools and what clinical use actually demands, comparing ChatGPT, a Quantitative Biomedical Research Center classifier and HistoWiz’s image-captioning model against pathology diagnostic standards.

  CISS-2025 proceedings (PDF, p. 48)

• Improving the quality of biomedical images using neural networks

  Petro Liashchynskyi, V. Fayerchuk, B. Halunka, V. Nadvynychnyy, L. Savanets

  YAISD Workshops (CEUR Workshop Proceedings)· 1 citations

  Neural approaches to enhancing biomedical image quality, addressing the noise and acquisition artefacts that degrade downstream classification and segmentation.

20247

• Synthesis of Convolutional Neural Network architectures for biomedical image classification

  Oleh Berezsky, Petro Liashchynskyi, Oleh Pitsun, Ivan Izonin

  Biomedical Signal Processing and Control (Elsevier)· 41 citations

  Automated synthesis of CNN architectures specialised for classifying biomedical images, rather than reusing architectures designed for natural-image benchmarks. The most cited of the journal papers.

  doi:10.1016/j.bspc.2024.106325

• Method of generative-adversarial networks searching architectures for biomedical images synthesis

  Oleh Berezsky, Petro Liashchynskyi

  Radio Electronics, Computer Science, Control· 5 citations

  Automates the design of GAN architectures rather than hand-tuning them. The synthesized cytological and histological images are then used to train the convolutional classifiers that support oncological diagnosis.

  doi:10.15588/1607-3274-2024-1-10

• Method and Software Tool for Generating Artificial Databases of Biomedical Images Based on Deep Neural Networks

  Oleh Berezsky, Petro Liashchynskyi, Oleh Pitsun, Grygoriy Melnyk

  arXiv / IDDM· 4 citations

  Develops a GAN-based method and accompanying software system for generating artificial biomedical images, including the data foundation and training-image module. The resulting synthetic database is compared against established public datasets.

  arXiv:2405.16119

• Synthesis of biomedical images based on generative intelligence tools

  Oleh Berezsky, Petro Liashchynskyi, Grygoriy Melnyk, Mykola Dombrovskyi, Mykhailo Berezkyi

  IDDM 2024 — Informatics & Data-Driven Medicine· 4 citations

  Applies generative models to biomedical image synthesis, continuing the group’s line of work on overcoming limited training data in medical imaging.

• Computer diagnostic systems: methods and tools

  Petro Liashchynskyi, Oleh Berezsky

  Ukrainian Journal of Information Technology· 1 citations

  A review of the methods and tools underpinning computer-aided diagnostic systems.

• Software tool for the classification and synthesis of biomedical images

  Petro Liashchynskyi

  Scientific Bulletin of UNFU (in Ukrainian)· 1 citations

  Biomedical image datasets are scarce, which holds back diagnostic tooling. This paper presents a modular, scalable software tool that generates synthetic but realistic medical images to augment the training data available to classifiers.

  doi:10.36930/40340415

• Synthesis of biomedical images based on deep neural networks

  Petro Liashchynskyi

  PhD dissertation — Lviv Polytechnic National University

  The doctoral thesis drawing this line of work together: generating biomedical images with deep networks to overcome the scarcity of real annotated medical data.

20233

• Deep network-based method and software for small sample biomedical image generation and classification

  Oleh Berezsky, Petro Liashchynskyi, Oleh Pitsun, Grygoriy Melnyk

  Radio Electronics, Computer Science, Control· 4 citations

  Addresses generating and classifying breast cancer histological images when only a small sample is available. Automating the diagnostic procedure saves time and removes the subjective element; the results are intended for cancer CAD systems.

  doi:10.15588/1607-3274-2023-4-8

• Analysis of metrics for GAN evaluation

  Petro Liashchynskyi, Pavlo Liashchynskyi

  Computer Systems and Information Technologies· 4 citations

  GAN training is unstable and hard to assess, and no consensus exists on which metrics genuinely reflect a model’s strengths and limitations. This paper reviews the proposed evaluation metrics and analyses what each actually measures when comparing generative models.

  doi:10.31891/csit-2023-4-6

• MLOps approach for automatic segmentation of biomedical images

  Oleh Berezsky, Oleh Pitsun, Grygoriy Melnyk, Yuriy Batko, Petro Liashchynskyi, Mykhailo Berezkyi

  IDDM 2023 — Informatics & Data-Driven Medicine· 3 citations

  Brings MLOps practice to medical image segmentation, treating model training and deployment as a reproducible pipeline rather than a one-off experiment.

20224

• Computational Intelligence in Medicine

  Oleh Berezsky, Oleh Pitsun, Petro Liashchynskyi, Bohdan Derysh, Natalia Batryn

  Intellectual Systems of Decision Making (Springer, LNDECT)· 16 citations

  A book chapter on the application of computational intelligence methods to medical decision-making and diagnostic support.

  doi:10.1007/978-3-031-16203-9_28

• Comparison of deep neural network learning algorithms for biomedical image processing

  Oleh Berezsky, Petro Liashchynskyi, Oleh Pitsun, Pavlo Liashchynskyi, Mykhailo Berezkyi

  IDDM 2022 — Informatics & Data-Driven Medicine· 14 citations

  An empirical comparison of training algorithms for deep networks on biomedical image processing tasks.

• Application of MLOps practices for biomedical image classification

  Oleh Berezsky, Oleh Pitsun, Grygoriy Melnyk, Yuriy Batko, Bohdan Derysh, Petro Liashchynskyi

  IDDM 2022 — Informatics & Data-Driven Medicine· 11 citations

  How MLOps practices — versioning, automated retraining, reproducible deployment — apply to biomedical image classifiers intended for clinical use.

• Design and Implementation of the Notification System: Project YASMIN

  Petro Liashchynskyi

  engrXiv (preprint)

  A systems-engineering preprint, separate from the biomedical imaging line: the design and implementation of a notification system.

  doi:10.31224/2678

20211

• Comparison of generative adversarial networks architectures for biomedical images synthesis

  Oleh Berezsky, Petro Liashchynskyi

  Applied Aspects of Information Technology· 9 citations

  Compares GAN architectures head-to-head for synthesising biomedical images, establishing which designs produce training data good enough to improve downstream classifiers.

  doi:10.15276/aait.03.2021.4

20192

• Grid Search, Random Search, Genetic Algorithm: A Big Comparison for NAS

  Petro Liashchynskyi, Pavlo Liashchynskyi

  arXiv· 1650 citations

  Compares the three most common hyperparameter optimisation algorithms — grid search, random search and a genetic algorithm — repurposed for neural architecture search, building CNNs and evaluating them on CIFAR-10 by both execution time and accuracy. By a wide margin the most cited work here.

  arXiv:1912.06059

• Synthesis of biomedical images based on generative adversarial networks

  Oleh Berezsky, Petro Liashchynskyi, Pavlo Liashchynskyi, A. Sukhovych, T. Dolynyuk

  Ukrainian Journal of Information Technology· 7 citations

  Acquiring biomedical images is expensive and slow. This work introduces BPCI2100, a database of 2,100 images of precancerous and cancerous breast tissue with accompanying patient and feature metadata, and develops a generator/discriminator GAN architecture to synthesise further images from it.

  doi:10.23939/ujit2019.01.035

20181

• GPU-based biomedical image processing

  Oleh Berezsky, Oleh Pitsun, Lesia Dubchak, Petro Liashchynskyi, Pavlo Liashchynskyi

  MEMSTECH 2018 (IEEE)· 11 citations

  GPU acceleration applied to biomedical image processing, addressing the throughput problem in analysing large histological and cytological image sets.

  doi:10.1109/MEMSTECH.2018.8365710

20171

• Intelligent system for automated microscopy of histological and cytological image analysis

  Oleh Berezsky, Oleh Pitsun, Petro Liashchynskyi, Grygoriy Melnyk

  Artificial Intelligence (in Ukrainian)· 14 citations

  An intelligent system for automated microscopy, analysing histological and cytological images — the earliest publication in this line of work.