These days, identifying medical deepfakes is crucial for preventing fraudulent activity, to avoid inaccurate diagnoses, as well as to uphold patient confidence. The massive increase in the production of realistic synthetic medical images presents significant challenges for clinical decision-making, highlighting the need for effective detection techniques. This proposed method offers a hybrid deepfake detection model which incorporates a lightweight Depth-Wise Convolution module in a Vision Transformer (DWConv-ViT) and a Fast Fourier Transform (FFT) module to improve feature extraction in the deepfake detection process. In contrast to conventional models which primarily use either frequency-based analysis or spatial analysis, our method integrates both feature types to increase resilience against malicious attacks. The proposed model was trained and tested using two datasets consisting of real knee X-ray images and GAN-generated osteoarthritis X-ray images. By utilizing both spatial and frequency-based details, our approach improves generalization and robustness against sophisticated deep-fake approaches. Therefore, this work helps to ensure the reliability and validity of medical diagnoses.