Foundation models, large-scale models pre-trained on vast and diverse datasets, effectively capture general knowledge and demonstrate strong robustness to a wide range of downstream tasks. However, the applicability of such foundation models to different domains and modalities has not been well studied. In this work, we introduce PathRadX, a cross-modal framework to evaluate the applicability of pathology foundation models to various classification tasks in radiology images. The framework integrate modality adaptation and task-specific classification strategies for efficient and effective cross-modal adaptation. Our findings highlight the cross-modal potential of pathology pre-trained models, demonstrating their applicability to radiology imaging tasks and paving the way for developing more efficient and accurate diagnostic tools.