Search papers, labs, and topics across Lattice.
To address the prohibitive deployment overhead of massive architectures in clinical imaging, LightMedSeg-ISLES establishes an ultra-compact 1.26-million-parameter pipeline for T1-weighted stroke lesion segmentation on the ISLES'26 benchmark. Evaluated on a 146-case held-out cohort, the model achieves a 0.618 mean Dice and 0.599 lesion-wise F1, retaining 97.5% of a 102.35M-parameter nnU-Net ResEnc-L's Dice score while surpassing it by 0.055 on lesion detection. These findings demonstrate that aggressive data augmentation and extended training schedules can bridge the capacity gap against models over 80脳 larger while reducing per-patch FLOPs by 4.7脳.
Shrinking a 102M-parameter nnU-Net by 81脳 costs barely 2.5% in segmentation Dice while actually boosting lesion-level detection F1 by over 5 points.
Large networks and ensembles often lead medical image segmentation challenges, but their storage and inference demands complicate deployment. We present LightMedSeg-ISLES, a 1.26-million-parameter pipeline for T1-weighted stroke lesion segmentation in ISLES'26. On a 146-case held-out cohort, flip test-time augmentation produces 0.618 mean Dice and 0.599 lesion-wise F1. A 102.35-million-parameter nnU-Net ResEnc-L produces 0.634 Dice and 0.544 lesion-wise F1 after size filtering. LightMedSeg therefore retains 97.5\% of nnU-Net's Dice with 81.4$\times$ fewer parameters while improving lesion-wise F1 by 0.055. Its four-pass TTA operating point requires 4.7$\times$ fewer FLOPs per standardized patch than nnU-Net. It also slightly exceeds filtered UNETR++ and nnFormer. Longer training and stronger augmentation add 0.0358 Dice without increasing capacity, establishing a strong single-checkpoint alternative to much larger models.
