DaX: New AI foundation model achieves best performance across 161 pathology tasks

A team of researchers (Bokai Zhao, Yiyang Zhang, Long Bai, Tai Ma, Hanqing Chao, Minfeng Xu) has introduced DaX, a pathology vision foundation model that leverages DINOv3-style self-supervised learning adapted for whole-slide histopathology. DaX is initialized from natural-image DINOv3 weights and incorporates several novel training strategies: continuous magnification training, cross-scale tissue views, orientation-agnostic and acquisition-robust augmentation, multi-input-size training, and Gram-anchored dense consistency. These designs enable the model to connect local cellular morphology with global tissue architecture while stabilizing dense token-level representations across input scales, making it robust to variations in magnification, staining, scanner type, and slide preparation.

To evaluate DaX, the team constructed a comprehensive WSI-level benchmark comprising 161 clinically meaningful tasks from 44 public datasets, covering 28,182 patients and 34,394 slides across four clinical domains and nine task categories. All models were evaluated under a fixed patient-level cross-validation protocol with fold-level statistical ranking. DaX achieved the highest mean performance across all tasks and consistently strong task-level ranking scores, with gains spanning diagnostic pathology, biomarker and molecular profiling, tissue/specimen context, and risk/response/prognosis. These results support DaX as a highly transferable visual encoder for computational pathology and provide a standardized evaluation framework for future pathology foundation models.