Adaptive Post-Processing Recovers Most of the Gap to nnU-Net v2 in Head and Neck GTV Segmentation: A Paired Three-Arm HECKTOR 2025 Benchmark
Year: 2026
Type: preprint
Abstract: Background Accurate delineation of the gross tumour volume (GTV) — primary tumour (GTVp) and nodal disease (GTVn) — on FDG-PET/CT is a critical step of head and neck radiotherapy planning. Comparisons between lightweight custom networks and the auto-configured nnU-Net v2 are usually reported as end-to-end pipelines, conflating the contribution of the network with that of the inference-time post-processing applied on top of it. We separated the two. Methods MiniUNet3D (custom 3D U-Net, 18.3 M parameters) and nnU-Net v2 ( 3d_fullres , 88.2 M parameters) were trained on the same 578 FDG-PET/CT cases (85/15 author-defined split of the HECKTOR 2025 Task 1 set, 8 centres) and evaluated on the same internal cohort. Three arms were compared pairwise: MiniUNet3D raw output at a fixed 0.5 threshold, MiniUNet3D with a locked adaptive post-processing pipeline, and nnU-Net v2. Comparisons used paired Wilcoxon tests with bootstrap confidence intervals, Bonferroni and Benjamini–Hochberg correction, and Cohen’s d; catastrophic failure (Dice <0.01) was compared with an exact McNemar test. Cases with an empty reference for a given target were excluded from that target’s analysis (n = 98 GTVp, n = 93 GTVn). Results With post-processing matched off, nnU-Net v2 was superior: median GTVp Dice 0.799 versus 0.592 (mean difference –0.244, 95 % CI –0.300 to –0.191; d = –0.88) and GTVn 0.774 versus 0.598 (d = –0.82). Post-processing raised MiniUNet3D to 0.800 (GTVp) and 0.738 (GTVn), recovering 79 % of that difference. Post-processed, MiniUNet3D matched nnU-Net v2 on GTVp Dice (p = 0.113) but remained inferior on nodal disease after Bonferroni correction (Dice p = 0.041; surface Dice p = 0.049). Catastrophic GTVp failures were 25/98 raw, 8/98 post-processed and 1/98 for nnU-Net v2 (McNemar p = 0.016). Inference took 34 s versus 78 s per case on the same GPU. Conclusions Post-processing recovered most, but not all, of the difference between the two models, and it did not confer robustness: an eight-fold higher rate of empty contours on small primaries persisted, which is the more consequential difference for planning safety. Pipeline comparisons reported without a post-processing ablation risk attributing to a network what post-processing supplied. Highlights Three-arm paired benchmark on HECKTOR 2025 Task 1: a lightweight custom 3D U-Net ( MiniUNet3D , 18.3 M parameters) with and without adaptive post-processing, against the auto-configured nnU-Net v2 (88 M parameters), on an identical training/validation split. With post-processing matched off in both arms, nnU-Net v2 is decisively better : median GTVp DSC 0.799 vs 0.592 (paired mean –0.244; Cohen’s d = –0.88; p < 0.001 after Bonferroni). Adaptive post-processing recovers most of that gap : it adds +0.192 DSC to MiniUNet3D on GTVp, 79 % of the 0.244 model-level difference. Post-processing is thus slightly smaller than the model gap itself, but three to four times larger than what remains of it afterwards (0.051 – 0.062). The gap is not fully closed. Even post-processed, nnU-Net v2 retains a small but Bonferroni-significant advantage on nodal disease (GTVn DSC p = 0.041; sDSC@2mm p = 0.049) and an eight-fold lower catastrophic failure rate (1/98 vs 8/98; McNemar exact p = 0.016). Practical implication: a team already committed to a lightweight custom network can recover most of what that choice costs at the inference stage, without retraining — but post-processing does not buy robustness against catastrophic failure on small primaries.
Source: medRxiv
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7204767972