Adapt innovative deep learning methods from breast cancer to Alzheimers disease

适应从乳腺癌到阿尔茨海默病的创新深度学习方法 摘要 阿尔茨海默病(AD)是痴呆症最常见的形式，受影响的美国人 预计到2050年将达到1340万。早期发现和治疗AD是预防非 可逆的致命的脑损伤。因此，来自神经成像方式的非侵入性标志物的发展 (如脑MRI)对AD的筛查和早期发现具有重要意义。在PI的Active R01奖中 (1R01EB032896-01)，研究重点是开发一条新的研究战略和技术路线 创新地分析乳腺癌图像以进行诊断、风险预测和分类。核心战略是 将医疗/临床智能融入数据驱动的深度学习建模。这项技术创新是 但不仅限于乳腺癌，还可以适应其他疾病，如阿尔茨海默病。因此，在这方面 补充建议，我们建议通过适应新的技术来开发我们的Active R01的AD重点 乳腺癌认知结果预测的创新，以发现早期和非侵入性成像 AD的生物标志物。这项补充研究的主要任务是利用大脑核磁共振成像建立深度学习模型 作为认知结果预测的输入，这是三个典型的分类问题中的一个 认知类：正常对照与轻度认知障碍与AD。我们提出了两个具体目标：1) 基于样本特征知识的深度课程学习，用于认知结果预测 2)从纵向脑MRI中学习知识，以提高AD的预测能力。我们将主要使用 公开可用的阿尔茨海默病神经成像倡议(ADNI)数据集。我们已经组装了一个多- 具有互补专业知识的纪律团队。拟议的研究将提供一种途径来翻译一些 在其他领域开发的创新技术以推进非侵入性成像生物标记物 AD的开发。该项目还将为PI的团队提供一个参与和 为AD相关的新研究做出贡献。

Adapt innovative deep learning methods from breast cancer to Alzheimer’s disease Abstract Alzheimer’s disease (AD) is the most common form of dementia, with the number of affected Americans expected to reach 13.4 million by the year 2050. Early detection and treatment of AD is critical to prevent non- reversible and fatal brain damage. Thus, development of non-invasive markers from neuroimaging modalities (e.g., brain MRI) is of great significance for screening and early detection of AD. In the PI’s active R01 award (1R01EB032896-01), the research focuses on developing a new line of research strategy and technical innovation to analyze breast cancer images for diagnosis, risk prediction, and triage. The core strategy is to incorporate medical/clinical intelligence into data-driven deep learning modeling. This technical innovation is however not limited to breast cancer, but can be adapted to other diseases, such as AD, as well. Thus, in this Supplement proposal, we propose to develop an AD focus of our active R01 by adapting the new technical innovation in breast cancer into cognitive outcome prediction for discovering early and no-invasive imaging biomarkers for AD. The main task of this Supplement study is to build deep learning models using brain MRIs as input for cognitive outcome prediction, which is formulated as a typical classification problem among three cognitive classes: Normal Control vs. Mild Cognitive Impairment vs. AD. We proposed two specific aims: 1) Deep curriculum learning informed by samples’ characteristic knowledge for cognitive outcome prediction and 2) Learning knowledge from longitudinal brain MRIs to improve prediction of AD. We will mainly use the publicly available Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset. We have assembled a multi- disciplinary team with complementary expertise. The proposed study will provide an avenue to translate some of the innovative techniques developed in other domains to advance non-invasive imaging biomarker development for AD. This project will also provide an opportunity for the PI’s team to get involved and contribute to AD-related new research.