Alzheimer’s disease (AD) is the most common cause of dementia, attributed to its characteristic symptoms of brain shrinkage and cell death. Onset of the diseases is thought to be due to abnormal buildup of amyloid-β (Aβ), a protein in the brain, creating plaques and leading to neurodegeneration. Similarly, Aβ buildup can cause cerebral amyloid angiopathy (CAA), a condition which significantly increases dementia and stroke risk.
Traditional methods for neuropathological assessments of diseases often rely on visual assessments of stained tissue sections. The semiquantitative nature of the scoring process can lead to large variation between pathologists and institutions. Artificial intelligence has immense potential to improve medical image analysis and data management.
With whole slide imaging (WSI) use of histological sections AI models can be used for histological object quantification, such as Aβ plaques, for both AD and CAA research. Professor Susanne van Veluw and Dr. Valentina Perosa have done exactly this with Aiforia Create. The researchers trained AI models to identify some of the most characteristic markers of AD and CAA, including Aβ plaques, iron and calcium deposition, reactive astrocytes, microglia, and fibrin extravasation; without the need to code or use dedicated hardware. We interviewed Susanne and Valentina to learn more about their recent publication and experience using Aiforia.
Interview with Susanne van Veluw, PhD (Assistant Professor) and Valentina Perosa, MD (Postdoctoral Research Fellow) at Massachusetts General Hospital, Department of Neurology
Describe the Alzheimer’s disease research project you used AI for.
Susanne and Valentina: The van Veluw lab used Aiforia to develop deep learning-based models that identify some of the most characteristic histopathological markers of Alzheimer’s disease (AD) and cerebral amyloid angiopathy (CAA). So far, the deep learning-based models were trained to differentially detect parenchymal amyloid β (Aβ)-plaques, vascular Aβ-deposition (CAA), iron and calcium deposition, reactive astrocytes, microglia, and fibrin extravasation. We have been using these measures in several projects, for example to explore the association between neuroinflammation and Aβ pathologies, and to confirm the relationship between iron-positive cells and CAA. Further work is ongoing in order to quantify more markers and investigate their crosstalk.
How is the AI model performing compared to using traditional methods?
Previously, we have mostly relied on semi quantitative scores to describe histopathological markers of AD and CAA. Our recent work showed that the deep learning-based measures are consistent with experts’ observations for CAA, Aβ-plaques, and iron-positive cells. Moreover, we performed a validation on the Aiforia platform, which showed good to excellent agreement between human observers and the AI.
What do you think makes Aiforia unique in Alzheimer’s disease research?
The workflow enabled by Aiforia is easy for researchers with pathological expertise to implement, without the need to directly interact with a code, and is customizable to the research questions. This characteristic makes AI in histopathology more accessible. More generally, AI provides objective continuous measures that allow the use of a greater variety of statistical models, such as linear mixed effects models. It can potentially also facilitate harmonization across datasets and centers, which opens new avenues for the relevance of histopathology in Alzheimer’s disease and CAA research.
How was your experience using the Aiforia software?
The Aiforia platform is easy to use and training a new AI-model is relatively straightforward. Clarification and help are readily provided by the Aiforia team when needed, so that the work can progress even if small obstacles should be encountered. The platform appears to be in constant development, allowing in the future to save even more time in the training and analysis process.