Medical Informatics

Research Article

DOI 10.25209/2079-3316-2026-17-3-3-56

The impact of morphological MRI features on the diagnosis of lymphovascular invasion in malignant breast neoplasms within a prototype clinical decision support system based on naive bayes nomograms

Ivan M. Skorobogach¹, Anton V. Vladzymyrskyy², Yury A. Vasiliev³, Ivan A. Blokhin⁴, Roman V. Reshetnikov⁵, Maria R. Kodenko⁶

1-6	Scientific and Practical Clinical Center for Diagnostics and Telemedicine Technologies of the Moscow Department of Health, Moscow, Russia
1	skorobogachim@zdrav.mos.ru

Abstract. Objective: To substantiate the role of morphological magnetic resonance imaging (m-MRI) features in hybrid morpho-radiomic models for the diagnosis of lymphovascular invasion (LVI) in malignant breast neoplasms and to develop a prototype of an explainable clinical decision support system (CDSS) based on naive Bayes nomograms with integrated SHAP analysis.

Materials and Methods: Data from 191 patients were analyzed: 13 m-MRI features and 12,390 radiomic (r-MRI) features extracted from the entire tumor volume (VOI^entire) and peritumoral tissues (VOI^{+2 mm}). The target variable was LVI. Predictor selection was performed via discretization using the Entropy-MDL method; classification was conducted using a naive Bayes algorithm with Laplace smoothing ($\alpha=1$ ). The CDSS prototype was implemented as a standalone single-page HTML application with integrated log-likelihood ratio and SHAP value calculation.

Results: Two radiomic naive Bayes (NB) models were developed: NB_VOI^entire (AUC=0.837) and NB_VOI^{+2 mm} (AUC-ROC=0.887). Multivariate logistic regression showed that peritumoral edema increased the odds of LVI by 5.66 times (OR 5.66, 95% CI: 2.27–14.94, p<0.001), and the tumor rim sign on DWI increased the odds by 4.05 times (OR 4.05, 95% CI: 1.63–10.47, p=0.003). Incorporating these m-MRI features into hybrid models significantly improved the AUC-ROC: an absolute increase of 7.6 percentage points (p=0.008) for the intratumoral model and 5.3 percentage points (p=0.049) for the extratumoral model.

Conclusion: The CDSS prototype implements the concept of explainable personalized diagnostics by combining naive Bayes nomograms with SHAP analysis. Incorporating m-MRI features of peritumoral edema and the DWI tumor rim sign into radiomic signatures is not merely supplementary but serves as a structural, framework-defining element that ensures the clinical robustness of the diagnostics. (Linked article texts in English and in Russian).

Keywords: clinical decision support system, explainable artificial intelligence, lymphovascular invasion, invasive malignant breast neoplasm, radiomics, naive Bayes, nomograms, Shapley value, MDL-Entropy

MSC-2020

93B45; 68T20, 92C50

For citation: Ivan M. Skorobogach, Anton V. Vladzymyrskyy, Yury A. Vasiliev, Ivan A. Blokhin, Roman V. Reshetnikov, Maria R. Kodenko. The impact of morphological MRI features on the diagnosis of lymphovascular invasion in malignant breast neoplasms within a prototype clinical decision support system based on naive bayes nomograms. Program Systems: Theory and Applications, 2026, 17:3, pp. 3–56. (in Engl. In Russ.). https://psta.psiras.ru/2026/3_3-56.

Full text of bilingual article (PDF): https://psta.psiras.ru/read/psta2026_3_3-56.pdf (Clicking on the flag in the header switches the page language).

The article was submitted 04.06.2026; approved after reviewing 29.06.2026; accepted for publication 29.06.2026; published online 22.07.2026.