<?xml version="1.0" encoding="UTF-8"?>
<article article-type="research-article" dtd-version="1.3" xml:lang="ru" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:noNamespaceSchemaLocation="https://metafora.rcsi.science/xsd_files/journal3.xsd">
  <front>
    <journal-meta>
      <journal-id journal-id-type="publisher-id">moitvivt</journal-id>
      <journal-title-group>
        <journal-title xml:lang="ru">Моделирование, оптимизация и информационные технологии</journal-title>
        <trans-title-group xml:lang="en">
          <trans-title>Modeling, Optimization and Information Technology</trans-title>
        </trans-title-group>
      </journal-title-group>
      <issn pub-type="epub">2310-6018</issn>
      <publisher>
        <publisher-name>Издательство</publisher-name>
      </publisher>
    </journal-meta>
    <article-meta>
      <article-id pub-id-type="doi">10.26102/2310-6018/2026.54.3.009</article-id>
      <article-id pub-id-type="custom" custom-type="elpub">2204</article-id>
      <title-group>
        <article-title xml:lang="ru">Вычислительный метод сегментации изображений на основе поля Дирихле и анализ асимптотической точности дискретизации пространственных регуляризаторов</article-title>
        <trans-title-group xml:lang="en">
          <trans-title>A computational method for image segmentation based on a Dirichlet field and an analysis of the asymptotic accuracy of spatial regularizer discretization</trans-title>
        </trans-title-group>
      </title-group>
      <contrib-group>
        <contrib contrib-type="author" corresp="yes">
          <contrib-id contrib-id-type="orcid">0000-0003-3651-7629</contrib-id>
          <name-alternatives>
            <name name-style="eastern" xml:lang="ru">
              <surname>Щетинин</surname>
              <given-names>Евгений Юрьевич</given-names>
            </name>
            <name name-style="western" xml:lang="en">
              <surname>Shchetinin</surname>
              <given-names>Evgeny Yuryevich</given-names>
            </name>
          </name-alternatives>
          <email>riviera-molto@mail.ru</email>
          <xref ref-type="aff">aff-1</xref>
        </contrib>
        <contrib contrib-type="author">
          <name-alternatives>
            <name name-style="eastern" xml:lang="ru">
              <surname>Шевчук</surname>
              <given-names>Андрей Андреевич</given-names>
            </name>
            <name name-style="western" xml:lang="en">
              <surname>Andreychuk</surname>
              <given-names>Andrey Andreevich</given-names>
            </name>
          </name-alternatives>
          <email>andreiluck11@yandex.ru</email>
          <xref ref-type="aff">aff-2</xref>
        </contrib>
      </contrib-group>
      <aff-alternatives id="aff-1">
        <aff xml:lang="ru">Севастопольский государственный университет</aff>
        <aff xml:lang="en">Sevastopol State University</aff>
      </aff-alternatives>
      <aff-alternatives id="aff-2">
        <aff xml:lang="ru">Севастопольский государственный университет</aff>
        <aff xml:lang="en">Sevastopol State University</aff>
      </aff-alternatives>
      <pub-date pub-type="epub">
        <day>01</day>
        <month>01</month>
        <year>2026</year>
      </pub-date>
      <volume>1</volume>
      <issue>1</issue>
      <elocation-id>10.26102/2310-6018/2026.54.3.009</elocation-id>
      <permissions>
        <copyright-statement>Copyright © Авторы, 2026</copyright-statement>
        <copyright-year>2026</copyright-year>
        <license license-type="creative-commons-attribution" xlink:href="https://creativecommons.org/licenses/by/4.0/">
          <license-p>This work is licensed under a Creative Commons Attribution 4.0 International License</license-p>
        </license>
      </permissions>
      <self-uri xlink:href="https://moitvivt.ru/ru/journal/article?id=2204"/>
      <abstract xml:lang="ru">
        <p>Предложен вычислительный метод семантической сегментации изображений с оценкой распределительной неопределенности на основе представления предсказания в виде поля распределений Дирихле. В отличие от подходов, требующих многократных стохастических прогонов при инференсе (MC-dropout) или усреднения по ансамблю независимых моделей, метод вычисляет карты неопределенности в замкнутой форме по параметрам поля Дирихле, предсказанным за один прямой проход нейросети. Метод формулируется как минимизация составного функционала, включающего ожидаемую логарифмическую функцию потерь (expected log-loss), KL-регуляризацию для управления концентрацией распределения и пространственное сглаживание, учитывающее локальные перепады интенсивности изображения (edge-aware). Для фиксированных гладких полей установлена асимптотическая точность дискретизации используемых пространственных регуляризаторов: дискретная энергия Дирихле аппроксимирует соответствующий непрерывный интеграл с погрешностью первого порядка по шагу сетки. Дополнительно введено формальное разложение общей неопределенности на эпистемическую и подтвержденную данными компоненты, которое может использоваться в дальнейшем при анализе поведения метода и построении расширений. Вычислительные эксперименты выполнены на трех наборах медицинских изображений (ACDC, Synapse, CHAOS) с 10 независимыми инициализациями. В основном сравнении с базовой моделью, обученной по кросс-энтропии, различия статистически значимы по инициализациям на всех датасетах; на ACDC дополнительно подтверждена значимость на уровне пациентов. Метод повышает качество сегментации и улучшает калибровку вероятностных оценок при накладных расходах порядка 17 %. В задаче детекции ошибок сегментации на уровне пикселей карта неопределенности достигает AUROC 0,891.</p>
      </abstract>
      <trans-abstract xml:lang="en">
        <p>A computational method for semantic image segmentation with distributional uncertainty estimation is proposed based on representing the prediction as a Dirichlet distribution field. Unlike approaches that require multiple stochastic inference runs (MC dropout) or averaging over an ensemble of independent models, the method computes uncertainty maps in closed form based on the Dirichlet field parameters predicted in a single forward pass of the neural network. The method is formulated as the minimization of a composite functional including the expected logarithmic loss function (expected log-loss), KL regularization for controlling the distribution concentration, and spatial smoothing that takes into account local image intensity variations (edge-aware). For fixed smooth fields, the asymptotic discretization accuracy of the spatial regularizers used is established: the discrete Dirichlet energy approximates the corresponding continuous integral with a first-order error over the grid step. Additionally, a formal decomposition of the overall uncertainty into epistemic and data-supported components was introduced, which can be used in further analysis of the method's behavior and the development of extensions. Computational experiments were performed on three medical image datasets (ACDC, Synapse, CHAOS) with 10 independent initializations. In the main comparison with the baseline model trained using cross-entropy, the differences are statistically significant across initializations on all datasets; for ACDC, significance at the patient level was further confirmed. The method improves segmentation quality and improves the calibration of probability estimates with an overhead of approximately 17 %. In the task of detecting pixel-level segmentation errors, the uncertainty map achieves an AUROC of 0.891.</p>
      </trans-abstract>
      <kwd-group xml:lang="ru">
        <kwd>сегментация изображений</kwd>
        <kwd>нейросетевые методы</kwd>
        <kwd>распределение Дирихле</kwd>
        <kwd>оценка неопределенности</kwd>
        <kwd>калибровка</kwd>
        <kwd>энергия Дирихле</kwd>
        <kwd>edge-aware регуляризация</kwd>
        <kwd>асимптотическая точность дискретизации</kwd>
      </kwd-group>
      <kwd-group xml:lang="en">
        <kwd>image segmentation</kwd>
        <kwd>neural network methods</kwd>
        <kwd>Dirichlet distribution</kwd>
        <kwd>uncertainty estimation</kwd>
        <kwd>calibration</kwd>
        <kwd>Dirichlet energy</kwd>
        <kwd>edge-aware regularization</kwd>
        <kwd>asymptotic sampling accuracy</kwd>
      </kwd-group>
      <funding-group>
        <funding-statement xml:lang="ru">Работа выполнена при финансовой поддержке Севастопольского государственного университета, проект 42-01-09/319/2025-1. </funding-statement>
        <funding-statement xml:lang="en">This work was supported by Sevastopol State University, project No. 42-01-09/319/2025-1.</funding-statement>
      </funding-group>
    </article-meta>
  </front>
  <back>
    <ref-list>
      <title>References</title>
      <ref id="cit1">
        <label>1</label>
        <mixed-citation xml:lang="ru">Begoli E., Bhattacharya T., Kusnezov D. The need for uncertainty quantification in machine-assisted medical decision making. Nature Machine Intelligence. 2019;1:20–23. https://doi.org/10.1038/s42256-018-0004-1</mixed-citation>
      </ref>
      <ref id="cit2">
        <label>2</label>
        <mixed-citation xml:lang="ru">Abdar M., Pourpanah F., Hussain S., et al. A review of uncertainty quantification in deep learning: Techniques, applications and challenges. Information Fusion. 2021;76:243–297. https://doi.org/10.1016/j.inffus.2021.05.008</mixed-citation>
      </ref>
      <ref id="cit3">
        <label>3</label>
        <mixed-citation xml:lang="ru">Kompa B., Snoek J., Beam A.L. Second opinion needed: communicating uncertainty in medical machine learning. npj Digital Medicine. 2021;4. https://doi.org/10.1038/s41746-020-00367-3</mixed-citation>
      </ref>
      <ref id="cit4">
        <label>4</label>
        <mixed-citation xml:lang="ru">Gal Y., Ghahramani Z. Dropout as a Bayesian approximation: Representing model uncertainty in deep learning. In: Proceedings of the 33rd International Conference on Machine Learning (ICML 2016), 19–24 June 2016, New York City, NY, USA. PMLR; 2016. P. 1050–1059.</mixed-citation>
      </ref>
      <ref id="cit5">
        <label>5</label>
        <mixed-citation xml:lang="ru">Lakshminarayanan B., Pritzel A., Blundell Ch. Simple and scalable predictive uncertainty estimation using deep ensembles. In: Advances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, 04–09 December 2017, Long Beach, CA, USA. 2017. P. 6402–6413.</mixed-citation>
      </ref>
      <ref id="cit6">
        <label>6</label>
        <mixed-citation xml:lang="ru">Maddox W.J., Izmailov P., Garipov T., et al. A simple baseline for Bayesian uncertainty in deep learning. In: Advances in Neural Information Processing Systems 32: Annual Conference on Neural Information Processing Systems 2019, NeurIPS 2019, 08–14 December 2019, Vancouver, BC, Canada. 2019. P. 13132–13143.</mixed-citation>
      </ref>
      <ref id="cit7">
        <label>7</label>
        <mixed-citation xml:lang="ru">Sensoy M., Kaplan L.M., Kandemir M. Evidential deep learning to quantify classification uncertainty. In: Advances in Neural Information Processing Systems 31: Annual Conference on Neural Information Processing Systems 2018, NeurIPS 2018, 03–08 December 2018, Montréal, Canada. 2018. P. 3183–3193.</mixed-citation>
      </ref>
      <ref id="cit8">
        <label>8</label>
        <mixed-citation xml:lang="ru">Malinin A., Gales M.J.F. Predictive uncertainty estimation via prior networks. In: Advances in Neural Information Processing Systems 31: Annual Conference on Neural Information Processing Systems 2018, NeurIPS 2018, 03–08 December 2018, Montréal, Canada. 2018. P. 7047–7058.</mixed-citation>
      </ref>
      <ref id="cit9">
        <label>9</label>
        <mixed-citation xml:lang="ru">Charpentier B., Zügner D., Günnemann S. Posterior network: Uncertainty estimation without OOD samples via density-based pseudo-counts. arXiv. URL: https://arxiv.org/abs/2006.09239 [Accessed 15th January 2026].</mixed-citation>
      </ref>
      <ref id="cit10">
        <label>10</label>
        <mixed-citation xml:lang="ru">Jungo A., Reyes M. Assessing reliability and challenges of uncertainty estimations for medical image segmentation. In: Medical Image Computing and Computer Assisted Intervention – MICCAI 2019: 22nd International Conference: Proceedings: Part II, 13–17 October 2019, Shenzhen, China. Cham: Springer; 2019. P. 48–56. https://doi.org/10.1007/978-3-030-32245-8_6</mixed-citation>
      </ref>
      <ref id="cit11">
        <label>11</label>
        <mixed-citation xml:lang="ru">Litjens G., Kooi Th., Bejnordi B.E., et al. A survey on deep learning in medical image analysis. Medical Image Analysis. 2017;42:60–88. https://doi.org/10.1016/j.media.2017.07.005</mixed-citation>
      </ref>
      <ref id="cit12">
        <label>12</label>
        <mixed-citation xml:lang="ru">Kingma D.P., Ba J. Adam: A method for stochastic optimization. arXiv. URL: https://arxiv.org/abs/1412.6980 [Accessed 15th January 2026].</mixed-citation>
      </ref>
      <ref id="cit13">
        <label>13</label>
        <mixed-citation xml:lang="ru">Chen T., Xu B., Zhang Ch., Guestrin C. Training deep nets with sublinear memory cost. arXiv. URL: https://arxiv.org/abs/1604.06174 [Accessed 15th January 2026].</mixed-citation>
      </ref>
      <ref id="cit14">
        <label>14</label>
        <mixed-citation xml:lang="ru">Micikevicius P., Narang Sh., Alben J., et al. Mixed precision training. arXiv. URL: https://arxiv.org/abs/1710.03740 [Accessed 18th January 2026].</mixed-citation>
      </ref>
      <ref id="cit15">
        <label>15</label>
        <mixed-citation xml:lang="ru">Bernard O., Lalande A., Zotti C., et al. Deep learning techniques for automatic MRI cardiac multi-structures segmentation and diagnosis: Is the problem solved? IEEE Transactions on Medical Imaging. 2018;37(11):2514–2525. https://doi.org/10.1109/TMI.2018.2837502</mixed-citation>
      </ref>
      <ref id="cit16">
        <label>16</label>
        <mixed-citation xml:lang="ru">Kavur A.E., Gezer N.S., Barış M., et al. CHAOS Challenge – combined (CT-MR) healthy abdominal organ segmentation. Medical Image Analysis. 2021;69. https://doi.org/10.1016/j.media.2020.101950</mixed-citation>
      </ref>
      <ref id="cit17">
        <label>17</label>
        <mixed-citation xml:lang="ru">Guo Ch., Pleiss G., Sun Y., Weinberger K.Q. On calibration of modern neural networks. In: Proceedings of the 34th International Conference on Machine Learning (ICML 2017), 06–11 August 2017, Sydney, NSW, Australia. PMLR; 2017. P. 1321–1330.</mixed-citation>
      </ref>
      <ref id="cit18">
        <label>18</label>
        <mixed-citation xml:lang="ru">Ronneberger O., Fischer Ph., Brox Th. U-Net: Convolutional networks for biomedical image segmentation. In: Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015: 18th International Conference: Proceedings: Part III, 05–09 October 2015, Munich, Germany. Cham: Springer; 2015. P. 234–241. https://doi.org/10.1007/978-3-319-24574-4_28</mixed-citation>
      </ref>
      <ref id="cit19">
        <label>19</label>
        <mixed-citation xml:lang="ru">He K., Zhang X., Ren Sh., Sun J. Deep residual learning for image recognition. In: 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 27–30 June 2016, Las Vegas, NV, USA. IEEE; 2016. P. 770–778. https://doi.org/10.1109/CVPR.2016.90</mixed-citation>
      </ref>
    </ref-list>
    <fn-group>
      <fn fn-type="conflict">
        <p>The authors declare that there are no conflicts of interest present.</p>
      </fn>
    </fn-group>
  </back>
</article>