Enhancing Breast Lesion Detection in Mammography through Teacher–Student Learning and Pseudo-Labeling
DOI:
https://doi.org/10.5281/zenodo.21912263Palavras-chave:
Breast lesion detection, Mammography analysis, Teacher–student learning, Pseudo-labeling, YOLO modelsResumo
This work investigates a teacher–student learning framework with pseudo-labeling for automated breast lesion detection in mammography using YOLO (You Only Look Once) object detection models. Motivated by the scarcity of reliable bounding-box annotations and the high variability of mammography datasets, the proposed approach leverages a previously trained teacher model to generate pseudo-labels that are subsequently used to train a student model. Experiments are conducted on the public VinDr-Mammo dataset from Vietnam, in conjunction with additional mammography datasets obtained from Kaggle, which are used for pseudo-label generation. The teacher model achieves a mAP@50 of 80.55% and precision of 80.73%. The student reaches a mAP@50 of 82.50% and precision of 88.42%. Overall, pseudo-labeling improves breast lesion detection in heterogeneous mammography scenarios, including it is possible to find for the code on Github.
Referências
CHEN, Yuanhong et al. BRAIxDet: Learning to detect malignant breast lesion with incomplete annotations. Medical image analysis, v. 96, p. 103192, 2024.
FATIMA, Maria; ZIA, Razia; USMANI, Irfan Ahmed. A hybrid framework integrating deep learning and XAI for breast cancer detection with stratified cross-validation. Systems and Soft Computing, p. 200412, 2025.
FIGUEIRAS, Hugo et al. Self-supervised learning for breast cancer detection: A review. Computers in Biology and Medicine, v. 198, p. 111245, 2025.
FIGUEIREDO, I. S., OLIVEIRA, J. P. G. de, GALINDO, J. M. da S., & VIEIRA, J. V. (2026). A progressive fine-tuning strategy for robust breast lesion detection across multiple mammography datasets. Caderno Pedagógico, 23(1), e22794.
KARACA AYDEMIR, Büşra Kübra et al. Detecting and classifying breast masses via YOLO-based deep learning. Neural Computing and Applications, v. 37, n. 17, p. 11555-11582, 2025.
KARIMI, Davood et al. Deep learning with noisy labels: Exploring techniques and remedies in medical image analysis. Medical image analysis, v. 65, p. 101759, 2020.
LI, Gang et al. Pseco: Pseudo labeling and consistency training for semi-supervised object detection. In: European Conference on Computer Vision. Cham: Springer Nature Switzerland, 2022. p. 457-472.
LI, Hengduo et al. Rethinking pseudo labels for semi-supervised object detection. In: Proceedings of the AAAI conference on artificial intelligence. 2022. p. 1314-1322.
LITJENS, Geert et al. A survey on deep learning in medical image analysis. Medical image analysis, v. 42, p. 60-88, 2017.
LIU, Yen-Cheng et al. Unbiased teacher for semi-supervised object detection. arXiv preprint arXiv:2102.09480, 2021.
LU, Liyun et al. Uncertainty-aware pseudo-label and consistency for semi-supervised medical image segmentation. Biomedical Signal Processing and Control, v. 79, p. 104203, 2023.
LU, Yunyao et al. Enhancing dual network based semi-supervised medical image segmentation with uncertainty-guided pseudo-labeling. Knowledge-Based Systems, p. 114454, 2025.
NGUYEN, Hieu T. et al. VinDr-Mammo: A large-scale benchmark dataset for computer-aided diagnosis in full-field digital mammography. Scientific Data, v. 10, n. 1, p. 277, 2023.
QI, Xuyin et al. Mediaug: Exploring visual augmentation in medical imaging. In: Annual Conference on Medical Image Understanding and Analysis. Cham: Springer Nature Switzerland, 2025. p. 218-232.
SHEHZADI, Tahira et al. Semi-supervised object detection: A survey on progress from cnn to transformer. Sensors (Basel, Switzerland), v. 26, n. 1, p. 310, 2026.
SILVA, Luiz Miguel Santos. Uma estratégia robusta de reavaliação de pseudo-rótulos para o algoritmo Self-Training. 2025.
SOHN, Kihyuk et al. Fixmatch: Simplifying semi-supervised learning with consistency and confidence. Advances in neural information processing systems, v. 33, p. 596-608, 2020.
SOHN, Kihyuk et al. A simple semi-supervised learning framework for object detection. arXiv preprint arXiv:2005.04757, 2020.
TEODORESCU, Vlad; OBREJA BRAȘOVEANU, Laura. Assessing the validity of k-fold cross-validation for model selection: Evidence from bankruptcy prediction using random forest and XGBoost. Computation, v. 13, n. 5, p. 127, 2025.
ZHANG, Ziqi et al. Robust Source-Free Domain Adaptation for Medical Image Segmentation based on Curriculum Learning. arXiv preprint arXiv:2510.08393, 2025.
