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

2026-04-27

AI Summary

This work addresses the capacity mismatch between frozen vision foundation models and lightweight decoders, which often leads to insufficient accuracy in segmenting infiltrative tumor boundaries. To overcome this limitation, the authors propose Dino-NestedUNet, a framework that integrates a pretrained DINOv3 encoder with a nested dense decoder. By leveraging dense connections and a multi-scale feature reuse mechanism, the model effectively fuses high-level semantic information with low-level morphological and textural details while enabling cross-scale feature calibration. Experimental results demonstrate that Dino-NestedUNet significantly outperforms UNet++ and Dino-UNet on the CHTN, OSU, and CAMELYON16 histopathology datasets. Furthermore, its strong zero-shot transfer performance on TIGER WSIBULK and OSU CRC underscores its exceptional cross-domain generalization capability.

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.

Problem

• computational pathology
• tumor bulk segmentation
• vision foundation models
• boundary fidelity
• domain shift

Innovation

• dense decoding
• vision foundation models
• tumor segmentation
• NestedUNet
• cross-domain generalization