Dino-NestedUNet: Unlocking Foundation Vision Encoders for Pathology Tumor Bulk Segmentation via Dense Decoding

TL;DR

This paper introduces Dino-NestedUNet, a dense decoding framework that enhances foundation vision encoder performance in pathology tumor segmentation, especially under domain shifts.

Contribution

It proposes a novel dense grid decoder that improves boundary accuracy by better utilizing pre-trained vision models in pathology segmentation.

Findings

Consistent improvements over UNet++ and Dino-UNet variants across multiple cohorts.

Effective zero-shot generalization to unseen datasets without fine-tuning.

Dense decoding enhances boundary fidelity in infiltrative tumor segmentation.

Abstract

Vision foundation models (VFMs), such as DINOv3, provide rich semantic representations that are promising for computational pathology. However, many current adaptations pair frozen VFMs with lightweight decoders, creating a capacity mismatch that often limits boundary fidelity for infiltrative tumor bulk segmentation. This paper presents Dino-NestedUNet, a framework that couples a pre-trained DINOv3 encoder with a Nested Dense Decoder. Instead of sparse skip connections and linear upsampling, the proposed decoder forms a dense grid of intermediate pathways to enable continuous feature reuse and multi-scale recalibration, aligning high-level semantics with low-level morphological textures during reconstruction. We evaluate Dino-NestedUNet on three histopathology cohorts (multi-center CHTN, institutional OSU, and CAMELYON16) and observe consistent improvements over UNet++ and standard Dino-UNet variants, particularly under cross-domain shift. To further assess external generalization, we perform zero-shot evaluation by training on CHTN and directly testing on unseen TIGER WSIBULK and OSU CRC cohorts without fine-tuning. These results suggest that dense decoding is a key ingredient for unlocking foundation encoders in boundary-sensitive pathology segmentation.