对比两种病理AI分析策略，指导有限资源下如何选择最佳方案。

A key issue in using AI for pathology diagnosis is what image information should be given to the AI and how limited analysis resources should be used. This study compares two ways of processing different types of images under limited resources. The first is WSI-AI, in which AI automatically compresses information from the whole slide image (WSI). The second is Diagnostic Field Selection (DFS)-AI in which an expert first selects several regions, magnifications, and comparisons needed for diagnosis, and the AI then analyzes those selected fields. To compare these two types, we built three image models: a rarely localized lesion, lesion detection in a nonuniform background, and a spatially continuous lesion. As a result, WSI-AI was better when a coarse view did not give enough information about lesion location in advance, as may occur with very small tumor foci in lymph nodes. In contrast, when a low-cost coarse view provided useful location information, DFS-AI was better in an intermediate range of limited resources. When searching for a specific object such as infectious organism in a nonuniform background, the relative value of DFS-AI increased. For spatially continuous lesions, WSI-AI found the presence of a lesion more easily as the lesion became larger. However, for complete characterization of the entire lesion, DFS-AI could be better because it used the continuous structure to select fields for analysis. In conclusion, both relative performance changes in a systematic way with location information from the coarse view, background heterogeneity, local context, and the spatial structure of the lesion. A practical strategy under limited resources is for physicians to choose between WSI-AI and DIS-AI according to the information structure of the diagnostic task.