Discrete and Continuous Models and Applied Computational ScienceDiscrete and Continuous Models and Applied Computational ScienceDiscrete and Continuous Models and Applied Computational Science2658-46702658-7149Peoples' Friendship University of Russia named after Patrice Lumumba (RUDN University)5192710.22363/2658-4670-2026-34-2-260-273JMNMUPLetters to the EditorПисьма в редакциюResearch ArticleEfficient Polyp Segmentation and Classification with YOLO11-SAM3 and LoRAЭффективная сегментация и классификация полипов с помощью YOLO11-SAM3 и LoRAhttps://orcid.org/0000-0003-3651-762916408533100O-8287-2017ShchetininEugene Yu.ЩетининЕ. Ю.

Doctor of Physical and Mathematical Sciences, Professor of the Department of Information Technology and Systems of the Sevastopol State University

riviera-molto@mail.ruhttps://orcid.org/0000-0002-1856-46438783969400B-8497-2016SevastianovLeonid A.СевастьяновЛ. А.

Doctor of Physical and Mathematical Sciences, Professor of the Department of Mathematical Modeling and Artificial Intelligence of the RUDN University

sevastianov-la@rudn.ruhttps://orcid.org/0000-0002-4643-327X55978487200TiutiunnikAnastasia A.ТютюнникА. А.

PhD in Physics and Mathematics, Associate Professor of the Department of Mathematical Modeling and Artificial Intelligence of the RUDN University

tyutyunnik-aa@rudn.ruSevastopol State UniversityСевастопольский государственный университетRUDN UniversityРоссийский университет дружбы народов15082026342VOL 34, NO2 (2026)ТОМ 34, №2 (2026)26027320082026Copyright ©; 2026, Shchetinin E.Y., Sevastianov L.A., Tiutiunnik A.A.Copyright ©; 2026, Щетинин Е.Ю., Севастьянов Л.А., Тютюнник А.А.2026Shchetinin E.Y., Sevastianov L.A., Tiutiunnik A.A.Щетинин Е.Ю., Севастьянов Л.А., Тютюнник А.А.https://creativecommons.org/licenses/by-nc/4.0https://journals.rudn.ru/miph/article/view/51927

Purpose To develop and experimentally evaluate an autonomous computer-aided diagnosis (CADx) system for colonoscopy that integrates polyp segmentation with binary histological classification (adenoma/non-adenoma) while maintaining computational efficiency for near real-time clinical deployment. Materials and Methods We propose a cascaded YOLO11-SAM3-LoRA architecture wherein a YOLO11-medium detector generates region proposals (bounding boxes), followed by SAM 3 performing box-prompted segmentation with parameter-efficient domain adaptation via Low-Rank Adaptation (LoRA) and a semantic anchoring mechanism based on text prompt ensembling. Histological classification is implemented through a dual-stream classifier that fuses region-of-interest (ROI) features extracted by EfficientNet-B0 with morphological mask descriptors. Training was conducted on the combined Kvasir-SEG and CVC-ClinicDB datasets (detection/segmentation) and CVC-HDClassif (classification). Generalization was evaluated on the external ETIS-Larib dataset under an external zero-shot protocol. Results On the primary test set, the system achieved a mean Dice coefficient (mDice) of 0.962 for segmentation, adenoma sensitivity of 92.3\%, and area under the receiver operating characteristic curve (AUROC) of 0.942 for binary classification. Under external zero-shot evaluation on ETIS-Larib, the system maintained segmentation quality (mDice = 0.884), indicating robustness to domain shift. End-to-end processing throughput reached 38.2 FPS on an NVIDIA A100 GPU. Conclusions The results demonstrate that the combination of cascaded localization, foundation model segmentation with LoRA adaptation, and text prompt ensembling achieves high segmentation quality and competitive CADx accuracy while preserving computational efficiency. Clinical deployment requires further prospective multicenter validation and error analysis stratified by lesion subtypes.

Цель Разработать и экспериментально оценить автономную систему компьютерной поддержки для колоноскопии, объединяющую сегментацию полипов и бинарную гистологическую классификацию (аденома/не-аденома) с сохранением вычислительной эффективности для near real-time применения. Материал и методы Предложена каскадная архитектура YOLO11--SAM3--LoRA: детектор YOLO11-medium формирует кандидаты (bounding boxes), после чего SAM3 выполняет box-prompted сегментацию с параметро-эффективной доменной адаптацией (LoRA) и механизмом семантического якорения на основе ансамбля текстовых промптов. Гистологическая классификация реализована двухпоточным классификатором, объединяющим признаки ROI (EfficientNet-B0) и морфологические дескрипторы маски. Обучение проведено на объединённом наборе Kvasir-SEG и CVC-ClinicDB (детекция/сегментация) и CVC-HDClassif (классификация). Переносимость оценивалась на внешнем датасете ETIS-Larib в режиме external zero-shot. Результаты На основной тестовой выборке достигнуты mDice = 0,962 для сегментации, чувствительность к аденомам (Sensitivity) = 92,3\% и AUROC = 0,942 в задаче бинарной классификации. Во внешнем тестировании на ETIS-Larib (zero-shot) система сохраняет качество сегментации (mDice = 0,884), что указывает на устойчивость к доменному сдвигу. Скорость end-to-end обработки составляет 38,2 FPS на NVIDIA A100. Заключение Результаты показывают, что сочетание каскадной локализации, foundation-сегментации с LoRA-адаптацией и ансамблирования текстовых промптов обеспечивает высокое качество сегментации и конкурентоспособную точность CADx при сохранении вычислительной эффективности. Для клинического внедрения необходима дальнейшая проспективная многоцентровая валидация и анализ ошибок по подтипам поражений.

polyp segmentationpolyp classificationcomputer–aided diagnosisCADxcolonoscopyYOLO11SAM 3LoRAprompt ensemblingdeep learningсегментация полиповклассификация полиповYOLO11SAM 3LoRAансамбль промптовCADCADxколоноскопияглубокое обучениеThe work was carried out with the financial support of the Sevastopol State University, project 42-01-09/319/2025-1J. C. van Rijn, J. B. Reitsma, J. Stoker, P. M. Bossuyt, S. J. van Deventer, and E. Dekker, “Polyp miss rate determined by tandem colonoscopy: a systematic review.,” The American journal of gastroenterology, vol. 101, no. 2, pp. 343-350, 2006. DOI: 10.1111/j.1572-0241.2006.00390.xD. K. Rex, C. Kahi, M. O’Brien, et al., “The American Society for Gastrointestinal Endoscopy PIVI (Preservation and Incorporation of Valuable Endoscopic Innovations) on real-time endoscopic assessment of the histology of diminutive colorectal polyps.,” Gastrointestinal endoscopy, vol. 73, no. 3, pp. 419-422, 2011. DOI: 10.1016/j.gie.2011.01.023O. Ronneberger, P. Fischer, and T. Brox, “U-Net: Convolutional Networks for Biomedical Image Segmentation,” in Medical Image Computing and Computer-Assisted Intervention - MICCAI 2015, ser. Lecture Notes in Computer Science, vol. 9351, Springer, Cham, 2015, pp. 234-241. DOI: 10.1007/978-3-319-24574-4_28Z. Zhou, M. Rahman Siddiquee, N. Tajbakhsh, and J. Liang, “UNet++: A Nested U-Net Architecture for Medical Image Segmentation,” in Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support. DLMIA ML-CDS 2018, ser. Lecture Notes in Computer Science, vol. 11045, Springer, Cham, 2018, pp. 3-11. DOI: 10.1007/978-3-030-00889-5_1F. Diakogiannis, F. Waldner, P. Caccetta, and C. Wu, “ResUNet-a: A deep learning framework for semantic segmentation of remotely sensed data,” ISPRS Journal of Photogrammetry and Remote Sensing, vol. 162, pp. 94-114, 2020. DOI: 10.1016/j.isprsjprs.2020.01.013D.-P. Fan and et al., “PraNet: Parallel Reverse Attention Network for Polyp Segmentation,” in Proceedings of the 2020 International Conference on Medical Image Computing and Computer-Assisted Intervention, ser. Lecture Notes in Computer Science, vol. 12266, Springer, Cham, 2020, pp. 263-273. DOI: 10.1007/978-3-030-59725-2_26B. Dong and et al., Polyp-PVT: Polyp segmentation with pyramid vision transformers, 2021.C. Wang, I. Yeh, and H.-Y. Liao, YOLOv9: Learning what you want to learn using programmable gradient information. 2024.A. Wang, H. Chen, L. Liu, K. Chen, Z. Lin, J. Han, and G. Ding, YOLOv10: Real-time end-to-end object detection. 2024.A. Kirillov and et al., Segment Anything, 2023.J. Ma, Y. He, F. Li, and et al., “Segment anything in medical images.,” Nature Communications, vol. 15, pp. 1-9, 2024. DOI: 10.1038/s41467-024-44824-zN. Ravi, V. Gabeur, Y. Hu, R. Hu, C. Ryali, T. Ma, and et al., SAM 2: Segment anything in images and videos. 2024.E. Shchetinin and A. Tiutiunnik, “On Segmentation of Polyps Using Segment Anything Model.,” Research result. Information technologies, vol. 10, no. 3, pp. 55-63, 2025. DOI: 10.18413/2518-1092-2025-10-3-0-5D. Jha, P. Smedsrud, M. Riegler, P. Halvorsen, T. de Lange, D. Johansen, and H. Johansen, “Kvasir-SEG: A segmented polyp dataset.,” in MultiMedia Modeling, ser. Lecture Notes in Computer Science, vol. 11962, Springer International Publishing, 2020, pp. 451-462. DOI: 10.1007/978-3-030-37734-2_37J. Bernal, F. Sánchez, G. Fernández-Esparrach, D. Gil, C. Rodríguez, and F. Vilariño, “WM- DOVA maps for accurate polyp highlighting in colonoscopy: Validation vs. saliency maps from physicians.,” Computerized Medical Imaging and Graphics, vol. 43, pp. 99-111, 2015. DOI: 10.1016/j.compmedimag.2015.02.007Y. Tudela, M. Majó, N. de la Fuente, and et al., “A complete benchmark for polyp detection, segmentation and classification in colonoscopy images.,” Frontiers in Oncology, vol. 14, pp. 1-19, 2024. DOI: doi.org/10.3389/fonc.2024.1417862J. Silva, A. Histace, O. Romain, X. Dray, and B. Granado, “Toward embedded detection of polyps in WCE images for early diagnosis of colorectal cancer.,” International Journal of Computer Assisted Radiology and Surgery, vol. 9, pp. 283-293, 2014. DOI: doi.org/10.1007/s11548-013-0926-3G. Jocher and J. Qiu, Ultralytics YOLO11 [Internet]. Version 11.0.0. GitHub, 2024.N. Carion, L. Gustafson, Y. Hu, and et al., SAM 3: Segment anything with concepts. 2025.E. Hu, Y. Shen, P. Wallis, Z. Allen-Zhu, and et al., LoRA: Low-rank adaptation of large language models. 2022.D. Jha, S. Ali, N. K. Tomar, H. D. Johansen, D. Johansen, J. Rittscher, M. A. Riegler, and P. Halvorsen, “Real-time polyp detection, localization and segmentation in colonoscopy using deep learning.,” IEEE Access, vol. 9, pp. 40 496-40 510, 2021. DOI: 10.1109/ACCESS.2021.3063716D. Jha, P. Smedsrud, M. Riegler, D. Johansen, T. de Lange, P. Halvorsen, and H. Johansen, “ResUNet++: An advanced architecture for medical image segmentation.,” in Proceedings of the IEEE International Symposium on Multimedia (ISM), ser. Lecture Notes in Computer Science,IEEE, 2019, pp. 225-230. DOI: 10.1109/ISM46123.2019.00049M. Mansoori, S. Shahabodini, J. Abouei, K. Plataniotis, and A. Mohammadi, Self-prompting polyp segmentation in colonoscopy using hybrid YOLO-SAM 2 model. 2024.D. Rex, D. Ahnen, J. Baron, and et al., “Serrated lesions of the colorectum: Review and recommendations from an expert panel.,” American Journal of Gastroenterology, vol. 107, no. 9, pp. 1315-1329, 2012. DOI: 10.1038/ajg.2012.161G. Ji, J. Liu, P. Xu, N. Barnes, F. Khan, S. Khan, and D. Fan, Frontiers in intelligent colonoscopy. 2025.