This study aimed to develop and evaluate a multimodal deep learning model that integrates panoramic radiographs and clinical data to improve the diagnosis of medication-related osteonecrosis of the jaw (MRONJ). Materials and Methods:The study included 705 panoramic radiographs (401 MRONJ and 304 Normal) collected from Ewha Womans University Mokdong and Seoul Hospitals. Radiographs were paired with patient information, including age, sex, body mass index (BMI), type of medication, duration, route of administration, dosage, and dental risk factors. Two deep learning models were compared: a ResNet-50 model using panoramic radiographs alone and the Modified ResNet-50 model incorporating both image and clinical features. Model performance was evaluated using accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC). Lesion localization and feature importance were analyzed using Gradient-weighted Class Activation Mapping (Grad-CAM) and SHapley Additive exPlanations (SHAP). Result: Significant differences in clinical characteristics were observed between the groups; the MRONJ group was older (74.5±10.0 vs. 68.6±12.7 years, P<0.01) and had a higher proportion of females (92.5% vs. 82.2%, P<0.01). In model performance, the Modified ResNet-50 model outperformed the PR-only model. Specifically, the Modified ResNet-50 model achieved an accuracy of 75.0%, a recall of 84.3%, and an AUC of 0.843, whereas the PR-only model showed lower performance with an accuracy of 72.1%, a recall of 81.4%, and an AUC of 0.720. SHAP analysis identified prolonged medication use and specific drug types (e.g., zoledronate, P<0.01) as influential predictors. Conclusion: The proposed multimodal deep learning framework demonstrated improved diagnostic performance and interpretability compared to radiograph-only models. This approach may serve as a practical and explainable AI-assisted tool for MRONJ diagnosis in routine dental practice.