Akademik Veri Yönetim Sistemi
9th International Artificial Intelligence and Data Processing Symposium, IDAP 2025, Malatya, Türkiye, 6 - 07 Eylül 2025, (Tam Metin Bildiri)
Automated analysis of lumbar spine MRI is essential for improving diagnostic consistency and enhancing clinical workflow efficiency in the evaluation of chronic low back pain (CLBP). In this study, we present a computer-aided diagnosis (CAD) framework designed to automate both the analysis of lumbar spine MRI scans and the generation of structured diagnostic reports. The system processes 3D DICOM MRI volumes by extracting mid-sagittal slices for the segmentation of vertebrae and intervertebral discs (IVDs), followed by a 3D cross-projection method to localize the corresponding axial slices. The SegResNet architecture is employed as segmentation model to delineate anatomical structures in both sagittal and axial views. From these segmentations, quantitative measurements of key spinal anatomy are extracted, enabling automated anatomical indices measurements, disorders detection, spinal stenosis severity grading, and evaluation of spinal alignment abnormalities. These assessments serve as the diagnostics information feed into the input prompt to large language model (LLM) for report generation. The proposed system leverages a novel retrievalaugmented generation (RAG) approach that integrates semantic retrieval and knowledge graph-based reasoning to generate detailed, level-specific diagnostic report. The system demonstrates high segmentation performance (Dice: 97.79% sagittal, 93.52% axial) and generates clinically coherent reports using AgenticRAG, achieving a BERT F1-score of 83.58%. These results highlight its effectiveness for accurate, level-specific diagnosis and streamlined clinical reporting.