The medial temporal lobe (MTL) is a hotspot for neuropathology, and measurements of MTL atrophy are often used as a biomarker for cognitive decline associated with neurodegenerative disease. Due to the aggregation of multiple proteinopathies in this region, the specific relationship of MTL atrophy to distinct neuropathologies is not well understood. Here, we develop two quantitative algorithms using deep learning to measure phosphorylated tau (p-tau) and TDP-43 (pTDP-43) pathology, which are both known to accumulate in the MTL and are associated with MTL neurodegeneration. We focus on these pathologies in the context of Alzheimer’s disease (AD) and limbic predominant age-related TDP-43 encephalopathy (LATE) and apply our deep learning algorithms to distinct histology sections, on which MTL subregions were digitally annotated. We demonstrate that both quantitative pathology measures show high agreement with expert visual ratings of pathology and discriminate well between pathology stages. In 140 cases with antemortem MR imaging, we compare the association of semi-quantitative and quantitative postmortem measures of these pathologies in the hippocampus with in vivo structural measures of the MTL and its subregions. We find widespread associations of p-tau pathology with MTL subregional structural measures, whereas pTDP-43 pathology had more limited associations with the hippocampus and entorhinal cortex. Quantitative measurements of p-tau pathology resulted in a significantly better model of antemortem structural measures than semi-quantitative ratings and showed strong associations with cortical thickness and volume. By providing a more granular measure of pathology, the quantitative p-tau measures also showed a significant negative association with structure in a severe AD subgroup where semi-quantitative ratings displayed a ceiling effect. Our findings demonstrate the advantages of using quantitative neuropathology to understand the relationship of pathology to structure, particularly for p-tau, and motivate the use of quantitative pathology measurements in future studies.

内侧颞叶 (MTL) 是神经病理学的热点，MTL 萎缩的测量通常用作与神经退行性疾病相关的认知能力下降的生物标志物。由于多种蛋白质病在该区域聚集，MTL 萎缩与不同神经病理学的具体关系尚不清楚。在这里，我们使用深度学习开发了两种定量算法来测量磷酸化 tau (p-tau) 和 TDP-43 (pTDP-43) 病理学，已知它们都会在 MTL 中积累并与 MTL 神经变性相关。我们专注于阿尔茨海默病 (AD) 和边缘主导的年龄相关 TDP-43 脑病 (LATE) 背景下的这些病理学，并将我们的深度学习算法应用于不同的组织学切片，并在其上对 MTL 子区域进行数字注释。我们证明，这两种定量病理学测量都与专家的病理视觉评级高度一致，并且能够很好地区分病理阶段。在 140 例死前 MR 成像病例中，我们比较了海马中这些病理的半定量和定量死后测量与 MTL 及其亚区域的体内结构测量的关联。我们发现 p-tau 病理学与 MTL 亚区域结构测量存在广泛关联，而 pTDP-43 病理学与海马和内嗅皮层的关联更为有限。 p-tau 病理学的定量测量产生了比半定量评级明显更好的生前结构测量模型，并且显示出与皮质厚度和体积的密切相关性。 通过提供更精细的病理学测量，定量 p-tau 测量还显示出与严重 AD 亚组结构的显着负相关，其中半定量评级显示天花板效应。我们的研究结果证明了使用定量神经病理学来理解病理学与结构的关系（尤其是 p-tau）的优势，并激励在未来的研究中使用定量病理学测量。