OBJECTIVE: To clinically validate a fully automated cartilage segmentation technique from quantitative double-echo steady-state (qDESS) MRI supporting simultaneous estimation of cartilage T2 and morphology. Here, we test whether laminar (superficial and deep layer) T2 results from convolutional neural network (CNN) segmentations are consistent with those from manual expert segmentations. MATERIALS AND METHODS: The 3D qDESS sequence was acquired using 3 T MRI (resolution: 0.3125 × 0.3125x1.5 mm) in both knees of 37 subjects with unilateral anterior cruciate ligament (ACL) injury and 48 uninjured controls. Automated femorotibial cartilage (FTJ) segmentation was based on a 2D U-Net. Laminar T2 and cartilage thickness across the FTJ) were compared between ACL-injured and contralateral knees, and between ACL-injured and control knees. Effect sizes of these differences were measured using non-parametric Cohen's d (d RESULT: Significant differences were observed only in deep T2, with longer T2 in ACL-injured knees than in contralateral and healthy control knees in most of the comparisons and with similar effect sizes for automated and manual segmentations (range d DISCUSSION: Fully-automated, CNN-based analysis showed similar sensitivity to differences in laminar cartilage T2 as manual segmentation, allowing automated qDESS-analyses to be applied to larger datasets.