This study presents an automated system for classifying breast cancer diagnosis and BI-RADS scores based on free-text mammography reports written in Romanian. Utilizing a newly collected and publicly available Romanian Dense Breast Mammography Collection (RDBMC) dataset, we define two classification tasks: (1) assessing the patient’s overall condition as healthy, benign, or malignant, and (2) predicting the severity of findings using the BI-RADS scoring system. To address the scarcity of research on Romanian medical reports, we experiment with both the original Romanian texts and their English translations. We evaluate four classifiers (Random Forest, Logistic Regression, Decision Trees, and Naive Bayes) combined with two text embedding methods (TF-IDF and Latent Semantic Indexing), incorporating patient age as an additional feature. Our results show that the Random Forest classifier with TF-IDF embeddings achieves the best performance for Romanian reports, with an accuracy of 80% in diagnosis classification and 73% in BI-RADS scoring, demonstrating robustness and potential clinical applicability. Although English translations yield slightly higher scores, differences are small, indicating that the system is effective across languages.