Akademik Veri Yönetim Sistemi
HASEKI TIP BULTENI-MEDICAL BULLETIN OF HASEKI, cilt.64, sa.2, ss.76-84, 2026 (ESCI, Scopus, TRDizin)
Aim: Idiopathic granulomatous mastitis (IGM) is a benign, chronic inflammatory disease of the breast, and its imaging findings may overlap with those of malignant non-mass enhancement (NME). This study aimed to investigate the performance of deep learning and machine learning models in differentiating IGM from malignant NME based on dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI). Methods: In this retrospective study conducted between January 2019 and March 2023, DCE-MRI findings of 30 patients with histopathologically confirmed IGM and of 33 patients with breast cancer presenting as NME were analyzed. The second dynamic phase of DCE-MRI (Dataset 1, 475 images) and the corresponding subtracted images (Dataset 2, 402 images) were used in this study. Datasets were sequentially split into 80% training and 20% testing sets to ensure a patient-level split. Image features were extracted using SqueezeNet and classified with a narrow neural network. Results: The mean age was significantly lower in the IGM group than in the NME group (41.3 +/- 11.3 vs. 52.2 +/- 11.4 years, p<0.001). For Dataset 1, the area under the curve was 0.997 in training and 0.870 in testing; for Dataset 2, the area under the curve was 0.998 in training and 0.807 in testing. Training accuracy was 0.984 (Dataset 1) and 0.978 (Dataset 2), whereas test accuracy was 0.811 (Dataset 1) and 0.704 (Dataset 2). Conclusion: The findings of this study suggest that deep learning shows significant promise for non-invasive differentiation of IGM from malignant NME on DCE-MRI, particularly in cases that are clinically indistinguishable.