NEURAL NETWORK MODEL FOR MEDICAL IMAGE SEGMENTATION WITH A LOSS FUNCTION BASED ON FRACTAL ANALYSIS
Computer Design Systems. Theory and Practice · 2026 · 8 (1) · 130-141
DOI: 10.23939/cds2026.01.130 ↗Kurz gefasst
Fachartikel in der Zeitschrift Computer Design Systems. Theory and Practice, online erschienen am 30.04.2026. Verfasst von Volodymyr Bereziuk und 1 weiteren Autor:innen. Die Arbeit nennt 14 Literaturangaben.
Abstract (Originalsprache)
This paper investigates the influence of the Grünwald–Letnikov fractional operator on the training process of brain MRI segmentation models. A integration model is proposed that combines the classical U-Net architecture with a physics-informed approach by incorporating a structural constraint into the loss function, describing the consistency between the fractally processed image and the predicted segmentation mask. The proposed approach ensures that high-frequency structural characteristics of the image are accounted for during model training. The implemented fractal physics-informed neural network demonstrated improved segmentation results compared to the standard U-Net architecture, particularly for degraded and noisy data. The inclusion of a fractal term in the loss function contributed to increased training stability and improved segmentation metrics. Analysis of the results confirmed the formation of sharper and geometrically consistent boundaries of segmented structures due to the use of a fractional derivative-based loss function component. At the same time, increased variability of the Dice coefficient on validation data was observed.
Zitieren
APA
Bereziuk, V., & Sokolovskyi, Y. (2026). NEURAL NETWORK MODEL FOR MEDICAL IMAGE SEGMENTATION WITH A LOSS FUNCTION BASED ON FRACTAL ANALYSIS. Computer Design Systems. Theory and Practice, 8(1), 130-141. https://doi.org/10.23939/cds2026.01.130
Vancouver
Bereziuk V, Sokolovskyi Y. NEURAL NETWORK MODEL FOR MEDICAL IMAGE SEGMENTATION WITH A LOSS FUNCTION BASED ON FRACTAL ANALYSIS. Computer Design Systems. Theory and Practice. 2026;8(1):130-141. doi:10.23939/cds2026.01.130
BibTeX
@article{Bereziuk2026,
author = {Bereziuk, Volodymyr and Sokolovskyi, Yaroslav},
title = {NEURAL NETWORK MODEL FOR MEDICAL IMAGE SEGMENTATION WITH A LOSS FUNCTION BASED ON FRACTAL ANALYSIS},
journal = {Computer Design Systems. Theory and Practice},
year = {2026},
volume = {8},
number = {1},
pages = {130--141},
doi = {10.23939/cds2026.01.130}
}
EUVIMED