Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
To address irregular lesion morphology and extreme scale variability in 3D CT pancreatic scans without compute-heavy models, the authors designed TRIUNE-Net, a compact 5.86M-parameter segmentation network. The architecture unifies stage-adaptive dilated convolutions, serial linear-deformable attention for non-convex boundaries, and lossless downsampling that prevents small-tumor feature loss typically caused by max pooling. Evaluated on MSD and NVD Pancreas datasets without external pre-training, it establishes a new state of the art, outperforming prior baselines by 6.6 points in sensitivity and 6.0 points in F1 score.
State-of-the-art 3D pancreatic tumor segmentation no longer demands massive architectures or pre-training: a tailored 5.86M-parameter model boosts lesion sensitivity by 6.6 points simply by replacing standard pooling with lossless downsampling and linear-deformable attention.
Pancreatic tumor segmentation in 3D CT volumes is challenged by extreme scale variability across both the pancreas and tumor, and highly irregular tumor morphology. While recent advances have pushed segmentation performance, existing methods do not explicitly address these challenges and come at the cost of excessive computational complexity, limiting their practicality in resource-constrained clinical environments. We propose TRIUNE-Net, a lightweight unified architecture that harmonizes scale, shape, and efficiency through three synergistic innovations. A multi-scale context aggregation module with stage-adaptive dilated convolutions enables the model to reason across the broad range of anatomical scales present in both organs. A serial linear-deformable attention mechanism combines large effective receptive fields with shapeadaptive deformable convolutions to capture irregular, non-convex tumor morphologies. Finally, an information-preserving downsampling module replaces conventional max pooling entirely, retaining all spatial information while adding negligible parameters, preventing small tumors from being discarded before they can be recognized. On both the MSD Pancreas and NVD Pancreas datasets, TRIUNE-Net achieves state-of-theart results with only 5.86 M parameters and no external pre-training, outperforming all baselines across all key tumor metrics. Specifically, it surpasses the next-best model by 0.45% in tumor Dice, 6.0 points in F1 score, 6.6 points in sensitivity, and 3.4 points in precision, simultaneously reflecting its ability to suppress both missed tumors and false alarms in clinically realistic conditions. Our code is available at: https://github.com/abdora-ai/TRIUNE-Net