[2407.15362] A Multimodal Knowledge-enhanced Whole-slide Pathology Foundation Model

View a PDF file from the paper entitled Full Diseases Foundation Foundation Model, by Yingxue Xu and 18 other authors
PDF HTML (experimental) view
a summary:Noticeable steps have been taken in mathematical pathology in the fair foundation model for the task that develops the performance of a wide range of clinical tasks in the direction of the river. Despite promising performance, there are still many challenges. First, the previous works resorted to either vision data only or as the image structure, with the disregard for pathology reports with more clinical information from pathologists and genetic expression definition files that, respectively, provide a distinctive knowledge of the user -use clinical applications. Second, the current progress in the FMS pathology is often focused on the level of correction, as the restricted context of training at the correction level fails to capture the entire slip patterns. Even FMS at the level of the modern slide is still struggling to provide the context of completely slipping to represent the correction. In this study, for the first time, we develop a basic model for pathology that includes three levels of methods: pathology slices, pathology reports, and genetic expression data, which led to 26,169 pairs at the slide level of 10,275 patients across 32 types of cancer, up to more than 116 million pathological patients. To take advantage of this CPATH data, we suggest a new new sliding model that injects the context of the multimedia slipping into the correction representation, which is called MSTAR before training. The proposed form in the pre -working function of CPATH, allowing FM pathology to obtain the entire context of slipping. As far as we know, this is the first attempt to integrate three methods in the context of fully slip to enhance FMS for pathology. To systematically evaluate the MSTAR capabilities, we have built the largest set of tumors standard, with 7 categories of oncology applications in 15 species of 97 therapeutic tasks.
The application date
From: Yingxue Xu [view email]
[v1]
Mon, 22 Jul 2024 04:09:27 UTC (14,386 KB)
[v2]
Monday, 5 August 2024 08:26:24 UTC (15,028 KB)
[v3]
Tuesday, 25 Mar 2025 08:49:58 UTC (19,100 KB)
2025-03-26 04:00:00