An accurate evaluation and detection of awareness after a severe brain injury is crucial to a patient's diagnosis, therapy, and end-of-life decisions. Misdiagnosis is frequent as behavior-based assessments often overlook subtle signs of consciousness. This study aimed to identify brain MRI characteristics of patients with residual consciousness after a severe brain injury and to develop a simple MRI-based scoring system according to the findings. We retrieved data from 128 patients and split them into a development or validation set. Structural brain MRIs were qualitatively assessed for lesions in 18 brain regions. We used logistic regression and support vector machine algorithms to first identify the most relevant brain regions predicting a patient's outcome in the development set. We next built a diagnostic MRI-based score and estimated its optimal diagnostic cut-off point. The classifiers were then tested on the validation set and their performance compared using the receiver operating characteristic curve. Relevant brain regions predicting negative outcome highly overlapped between both classifiers and included the left mesencephalon, right basal ganglia, right thalamus, right parietal cortex, and left frontal cortex. The support vector machine classifier showed higher accuracy (0.93, 95% CI: 0.81-0.96) and specificity (0.97, 95% CI: 0.85-1) than logistic regression (accuracy: 0.87, 95% CI: 0.73 - 0.95; specificity: 0.90, 95% CI: 0.75-0.97), but equal sensitivity (0.67, 95% CI: 0.24-0.94 and 0.22-0.96, respectively) for distinguishing patients with and without residual consciousness. The novel MRI-based score assessing brain lesions in patients with disorders of consciousness accurately detects patients with residual consciousness. It could complement valuably behavioral evaluation as it is time-efficient and requires only conventional MRI.