Radiology reports often contain important diagnostic information but are written in semi-structured formats, making data extraction difficult. At the Kara Clinic in Oran, Algeria, radiologists face challenges in manually retrieving key details. This study presents an automated approach using Natural Language Processing (NLP) to extract essential findings and generate structured summaries. The system includes pre-processing, pseudonymization, and entity extraction using a domain-adapted model (Jean-Baptiste/camembert-ner), applied to pre-segmented reports. The quality of the generated summaries is evaluated using BLEU and ROUGE scores. Extracted keywords will also support future work on feature creation for training models in abdominal image segmentation. The goal is to implement a consistent and flexible pipeline that consistently delivers accurate results and can be confidently used by healthcare professionals to support faster and more efficient clinical decision-making.