Fig. 1: The primary strength and overall framework of Pathology-NAS. | npj Digital Medicine

Fig. 1: The primary strength and overall framework of Pathology-NAS.

From: Large language models driven neural architecture search for universal and lightweight disease diagnosis on histopathology slide images

Fig. 1

a Pathology NAS is trained on a large-scale corpus of medical image datasets. PrevIoUs specific disease solutions require to design customized model for each disease, lacking enough generalization capabilities. While universal foundation models are trained on data covering numerous anatomical structures, high-cost customization for complex cases is still required. Compared with disease-specific methods and existing universal foundation models, Pathology-NAS holds advantages in automated versatility and lightweight model design. b Overview of Pathology-NAS, a universally medical image analysis framework driven by LLM-assisted neural architecture search. Pathology-NAS significantly benefits from large-scale generic supernet pretraining, neural architecture fine-tuning and validation on diverse pathology datasets, and LLM-assisted architecture search.

Back to article page