Background: Delay in diagnosing metastatic epidural spinal cord compression (MESCC) adversely impacts clinical outcomes. High-grade MESCC is frequently overlooked on routine staging CT scans. We aim to assess the potential of our deep learning model (DLM) in detecting high-grade MESCC and reducing diagnostic delays. Methods: This retrospective review analyzed 140 patients with surgically treated MESCC between C7 and L2 during 2015–2022. An experienced radiologist (serving as the reference standard), a consultant spine surgeon, and the DLM independently classified staging CT scans into high-grade MESCC or not. The findings were compared to original radiologist (OR) reports; inter-rater agreement was assessed. Diagnostic delay referred to the number of days elapsed from CT to diagnostic MRI scan. Results: Overall, 95/140 (67.8%) patients had preoperative CT scans. High-grade MESCC was identified in 84/95 (88.4%) of the scans by the radiologist (reference standard), but in only 32/95 (33.7%) of the preoperative scans reported by the OR. There was almost perfect agreement between the radiologist and the surgeon (kappa = 0.947, 95% CI = 0.893–1.000) (p < 0.001), and between the radiologist and the DLM (kappa = 0.891, 95% CI = 0.816–0.967) (p < 0.001). In contrast, inter-observer agreement between the OR and all other readers was slight (kappa range = 0.022–0.125). Diagnostic delay was potentially reduced by 20 ± 28 (range = 1–131) days. Conclusions: The original radiologist reports frequently missed high-grade MESCC in staging CT. Our DLM for CT diagnosis of high-grade MESCC showed almost perfect inter-rater agreement with two experienced reviewers. This study is the first to demonstrate that the DLM could help reduce diagnostic delays. Further prospective research is required to understand its precise role in improving the early diagnosis/treatment of MESCC.