Medical images are essential for diagnosing and determining certain diseases. Securing transmitted medical images is essential where insecure networks negatively influence important patient data. Several medical image-securing methods have succeeded, but their robustness against complex attacks has been neglected, which motivated researchers to strengthen medical image-securing methods against complex attacks. To address this issue, we proposed an innovative zero-watermarking method based on DarkNet-53, inspired by the Fibonacci Q-matrix and stationary wavelet transform (SWT), to secure medical images. A single-level SWT transforms the host image and constructs the detailed coefficients. We feed the constructed coefficients into a pre-trained DarkNet-53 model to extract a robust feature vector. The pre-trained model, when integrated with SWT in an amalgamation domain, accurately captures more native and local image features. We encrypt the binary watermark (BW) using the Fibonacci Q matrix cryptosystem and fuse it with the featured image through an XOR operation to enhance security. Incorporating this cryptosystem into the framework significantly strengthens its resistance to various counterattacks. To achieve high robustness, this cryptosystem scrambles both the extracted feature matrix and BW bits. The experimental results are very encouraging and prohibit the efficiency of the proposed method in terms of robustness and invisibility. The extracted watermark appears to be original with optimal BER and NC values. In nearly all attack scenarios, the BER values were close to zero and the NC values were close to one. Compared to previous methods, the proposed method provides significant improvements in terms of robustness and invisibility.