EDI — Elektronikas un datorzinātņu institūts
Atpakaļ uz personālu

Romans Maļiks

Mg. sc. ing.

Pētnieks

Romans Maļiks EDI strādā kopš 2016. gada. Absolvējis Rīgas Tehniskās universitātes Elektronikas un telekomunikāciju fakultāti, kur 2018. gadā ieguvis inženierzinātņu maģistra grādu elektronikā; vienu semestri studējis Erasmus+ ietvaros Karaliskajā Tehnoloģijas institūtā Stokholmā. Šobrīd turpina studijas doktorantūrā elektronikas programmā. Pētniecības intereses aptver mākslīgo intelektu, mašīnmācīšanos un datu zinātni, integrālo shēmu projektēšanu un iegultās sistēmas. Fokusējas uz šo jomu saskarsmes punktiem — no algoritmu līmeņa līdz aparatūras īstenošanai.

Būtisks sasniegums ir pilna integrālās mikroshēmas izstrādes cikla apguve, izmantojot BiCMOS (SiGe:C) tehnoloģiju — no shēmas projektēšanas līdz mikroshēmas pasūtīšanai ražotājam. Papildus tam ir uzkrāta praktiska pieredze STM32 mikrokontrolieru iegultās sistēmās. Programmē Python un C valodās.

Projekti

1 rezultāts

Publikācijas

  • 2025

    Virtual Staining From Optical Coherence Tomography to Hematoxylin and Eosin Stained Skin Tumor Samples in Pets

    Daira Viškere, Mindaugas Tamošiūnas, Romans Maļiks, Diāna Dupļevska, Ilze Matīse-van Houtana, Roberts Kadiķis, Blaž Cugmas

    31st Annual International Scientific Conference "Research for Rural Development 2025

    Kopsavilkums

    Histopathological biopsies with hematoxylin and eosin–stained (H&E) samples are gold standard in veterinary oncology. Since biopsy sampling is an invasive process and requires sedation of the animal, as well as sample preparation and obtaining results is time-consuming, other effective solutions are being sought. Optical coherence tomography (OCT) is known as non-invasive imaging technology that produces detail, cross-sectional images of tissues at a microscopic level and provide real-time images. In our research we trained neural network to produce virtually stained histopathological images from OCT images taken from lipomas, soft tissue sarcomas and mast cell tumors from dogs and cats. As there are no previous research done on recognizing different normal and tumor structures in veterinary skin tumor samples, we found that tumor borders, hair follicles can be recognized using noninvasive OCT. As reference for virtually stained image production H&E sections were used. Multiple 1x1 mm sections of input images were taken from different (normal and tumor tissue) regions. We trained our neural network using paired dataset and got first virtually stained H&E images. By modifying these algorithms for transfer learning we could improve future images making more important histological structures to be recognized and improving image quality allowing veterinarians make tumor diagnostics much faster.

  • 2025

    Raman spectral band imaging for the diagnostics and classification of canine and feline cutaneous tumors(atveras jaunā logā)

    Tamošiūnas Mindaugas, Maciulevičius Martynas, Maļiks Romans, Dupļevska Diāna, Viškere Daira, Matīse-van Houtana Ilze, Kadiķis Roberts, Cugmas Blaž, Raišutis Renaldas

    Veterinary Quarterly (Vet Q), Sēj. 45 (1), lpp. 1–17, Taylor & Francis

    Kopsavilkums

    This study introduces Raman imaging technique for diagnosing skin cancer in veterinary oncology patients (dogs and cats). Initially, Raman spectral bands (with specificity to certain molecular structures and functional groups) were identified in formalin-fixed samples of mast cell tumors and soft tissue sarcomas, obtained through routine veterinary biopsy submissions. Then, a custom-built Raman macro-imaging system featuring an intensified CCD camera (iXon Ultra 888, Andor, UK), tunable narrow-band Semrock (USA) optical filter compartment was used to map the spectral features at 1437 cm−1 and 1655 cm−1 in ex vivo tissue. This approach enabled wide-field (cm2), rapid (within seconds), and safe (< 400 mW/cm2) imaging conditions, supporting accurate diagnosis of tissue state. The findings indicate that machine learning classifiers–particularly support vector machine (SVM) and decision tree (DT)–effectively distinguished between soft tissue sarcoma, mastocytoma and benign tissues using Raman spectral band imaging data. Additionally, combining Raman macro-imaging with residual near-infrared (NIR) autofluorescence as a bimodal imaging technique enhanced diagnostic performance, reaching 85–95% in accuracy, sensitivity, specificity, and precision–even with a single spectral band (1437 cm−1 or 1655 cm−1). In conclusion, the proposed bi-modal imaging is a pioneering method for veterinary oncology science, offering to improve the diagnostic accuracy of malignant tumors. © 2025 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group.

    Indeksēts Scopus / WoS · Scopus procentile: 98

  • 2024

    Interpreting microscopic structures in virtually stained histological sections for veterinary oncology applications

    Diāna Dupļevska, Romans Maļiks, Mindaugas Tamošiūnas, Mikus Melderis, Daira Viškere, Blaž Cugmas, Roberts Kadiķis, Ilze Matīse-van

    SPIE BiOS 2024, Advanced Biomedical and Clinical Diagnostic and Surgical Guidance Systems XXII, Sēj. 128310, SPIE

    Kopsavilkums

    We employed a Pix2Pix generative adversarial network to translate the multispectral fluorescence images into colored brightfield representations resembling H&E staining. The model underwent training using 512x512 pixel paired image patches, with a manually stained image serving as the reference and the fluorescence images serving as the processing input. The baseline model, without any modifications, did not achieve high microscopic accuracy, manifesting incorrect color attribution to various biological structures and the addition or removal of image features. However, through the substitution of simple convolutions with Dense convolution units in the U-Net Generator, we observed an increase in the similarity of microscopic structures and the color balance between the paired images. The resulting improvements underscore the potential utility of virtual staining in histopathological analysis for veterinary oncology applications.

    Indeksēts Scopus / WoS · Scopus procentile: 23 · SNIP: 0.22

  • 2023

    Wide-field Raman spectral band imaging of tumor lesions in veterinary medicine(atveras jaunā logā)

    Mindaugas Tamošiūnas, Roberts Kadiķis, Mikus Melderis, Romāns Maļiks, Diāna Duplevska, Daira Viškere, Ilze Matīse-van Houtana, Blaž Cugmas

    Translational Biophotonics: Diagnostics and Therapeutics III, Sēj. 12627, lpp. 295-302, SPIE

    Kopsavilkums

    Skin and subcutaneous tumors are common in companion animals, that can be difficult to diagnose and treat. Raman spectroscopy shows high diagnostic accuracy in identifying malignant tumors and benign lipomas in dogs and cats. However, the traditional single-point raster scanning approach is not ideal for a large-field-of-view Raman imaging due to its time-consuming nature when scanning areas larger than a square centimeter. Additionally, focusing the excitation spot can lead to high levels of light fluency (J/cm2), potentially causing damage to tissue biomolecules. Furthermore, the resulting raster-scan image often lacks sufficient spatial resolution to effectively compare it with tissue morphology findings. In this study, we focused on implementing EMCCD camera-based Raman imaging to accurately capture Raman spectral band signatures and overcome autofluorescence interference in veterinary cancer samples ex vivo. By utilizing the tunable band-pass filters set-up, our system enables large-field-of-view imaging of specific Raman bands, such as the 1437 cm-1 band or 1652 cm-1 band in biological tissue, proposing a more efficient, accurate and safe approach for Raman imaging in the veterinary field.

    Indeksēts Scopus / WoS · Scopus procentile: 8 · SNIP: 0.235

  • 2023

    Evaluation of Materials and Structures with a Multistatic Ultra-Wideband Impulse Radar: A Concept Validation(atveras jaunā logā)

    Gatis Gaigals, Romans Maliks, Vladimir Aristov, Rolands Savelis, Janis Simanovics, Eduards Lobanovs, Haralds Egliens, Dans Laksis, Kristaps Maris Greitans, Modris Greitans.

    Applied Sciences, Sēj. 13 (3), lpp. 1636, MDPI

    Kopsavilkums

    This document describes the results of the study contributing to the methods and tools applicable in plastic waste sorting systems that exploit the multistatic ultra-wideband impulse radar enforced with a deep learning signal processing back-end. The novelty of the research is the use of synthetic data for the development of a trained neural network before real data are available, and the use of a multistatic radar for the improvement of the training data set. The study results are described in multiple publications; the current paper shows the applicability of the described approach. The main results are as follows: a monostatic impulse radar can be used for the determination of material properties, such as thickness, dielectric permittivity, and losses, with limited accuracy; multistatic radar configuration increases the accuracy of the material property estimation; an open source finite difference time domain simulator can be used to simulate electromagnetic wave propagation in dielectric structures in order to generate synthetic data for development of optimized artificial neuron network structures used for the estimation of dielectric material properties, and the developed network can successfully be used for multistatic radar data processing.

    Scopus procentile: 73 · SNIP: 1.026