The aim of this study is to present a deep learning (DL) algorithm for accurate C3 paravertebral space muscle delineation in magnetic resonance (MR) images of patients with oral squamous cell carcinoma (OSCC). This study tries to enhance the methodologies employed by medical professionals in radiomics by incorporating operator-independent segmentation techniques. Such approaches are essential for accurately identifying target regions and developing reliable predictive models. We investigated a deep learning model C-ENet, using head and neck MR scans from 146 patients, either pre-surgical or during follow-up. The model’s segmentation performance was assessed by comparing its output to expert-defined ground truth annotations. Our findings indicate that the C-ENet model can achieve effective segmentation of C3 paravertebral space muscles, reaching a Dice Similarity Coefficient (DSC) of \(67.41\%\) . In summary, the study confirms that deep learning can be successfully utilized for fast, radiologist-free segmentation of the C3 paravertebral region, enabling consistent, user-independent outputs for downstream radiomics applications.