N. Yu. Ilyasova1, D.V. Nekrasova2, E.V. Zarov3, N.S. Demin4, S.S. Pervushkin5

1–4 Samara National Research University (Samara, Russia)
1,4 Image Processing Systems Institute, NRC «Kurchatov Institute» (Moscow, Russia)
5 Samara State Medical University Samara (Samara, Russia)
1 ilyasova.nata@gmail.com, 2 dashanek007@gmail.com, 3 zarov.gcg@gmail.com, 4 volfgunus@gmail.com,
5 Sergey.pervushkin@gmail.com

This study presents a comparative analysis of two approaches for the automatic diagnosis of osteoporosis from lumbar spine radiographs: a purely neural network method and a hybrid approach combining deep learning with texture feature analysis. The relevance of this research is due to the growing prevalence of osteoporosis, which affects one in three women and one in four men over the age of 50, and the need to enhance diagnostic accuracy and reduce subjectivity in clinical decision-making. The study aims to evaluate the effectiveness of both approaches by analyzing their segmentation and classifi-cation performance. The neural network method relies on deep learning architectures such as DeepLabV3+ for segmentation and ResNet-34 for classification, while the hybrid method integrates traditional texture feature extraction (GLCM, LBP, and wavelet transforms) with the Random Forest classifier. The results demonstrate that the neural network approach achieves an accuracy of 87.42%, while the hybrid method, despite a lower overall accuracy of 78%, provides higher recall (86%) for positive osteoporosis cases, minimizing false negatives. These findings indicate that deep learning offers high precision, whereas hybrid models provide better interpretability and sensitivity. The proposed approaches con-tribute to the development of automated diagnostic tools, enhancing the efficiency and reliability of osteoporosis detection in clinical practice.

Ilyasova N.Yu., Nekrasova D.V., Zarov E.V., Demin N.S., Pervushkin S.S. Neural network and hybrid technologies in osteoporosis diagnostics: analysis of x-ray images of the spine. Biomedicine Radioengineering. 2025. V. 28. № 5. P. 141–144. DOI: https:// doi.org/10.18127/j15604136-202505-28 (In Russian)

1. Verbovoj A.F., Pashenceva A.V., Sharonova L.A. Osteoporoz: sovremennoe sostoyanie. Terapevticheskij arhiv. 2017. T. 89, No 5. S. 90–97. DOI 10.17116/terarkh201789590-97. EDN YSYUSP (In Russian).
2. Klimova ZH.A., Zaft A.A., Zaft V.B. Sovremennaya laboratornaya diagnostika osteoporoza. Mezhdunarodnyj endokrinologicheskij zhurnal. 2014. S. 1–10 (In Russian).
3. Kruz A.S., Lins H.K., Medejros R.V.A. i dr. Iskusstvennyj intellekt po identifikacii grupp riska osteoporoza, obshchij obzor. Biomedicinskaya inzheneriya onlajn. 2018. T. 17. S. 1–17 (In Russian).
4. Gajdel' A.V., Pervushkin S.S. Issledovanie teksturnyh osobennostej dlya diagnostiki zabolevanij kostnoj tkani s pomoshch'yu rentgenogramm. Komp'yuternaya optika. 2013. T. 37. № 1. S. 113–120 (In Russian).
5. Ren S., He K., Girshik R., San D. Bolee bystryj R-CNN: na puti k obnaruzheniyu ob"ektov v real'nom vremeni s pomoshch'yu Region Proposal Networks. Trudy IEEE po analizu obrazov i mashinnomu intellektu. 2016. T. 39. № 6. S. 1137–1149(In Russian).
6. Chen L.S., Papandreu G., Shroff F., Adam H. Pereosmyslenie stroevoj svertki dlya semanticheskoj segmentacii izobrazhenij. Komp'yuternoe zrenie i raspoznavanie obrazov. 2017. DOI: https://doi.org/10.48550/arXiv.1506.01497 (In Russian).
7. Haralik R.M., Shanmugam K., Dinshtejn I. Teksturnye osobennosti dlya klassifikacii izobrazhenij. IEEE Transactions on Systems, Man, and Cybernetics. 1973. No 6. P. 610–621 (In Russian).
8. Huang D., Shan' K., Ardabilyan M. i dr. Lokal'nye binarnye patterny i ih primenenie k analizu izobrazhenij lica: obzor. IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews). 2011. T. 41, № 6. S. 765–781 (In Russian).