Deep learning tools for the automated analysis of hematopathology whole slide images and the development of prognostic algorithms for hematopoietic stem cell transplant recipients

用于自动分析血液病理学全幻灯片图像和开发造血干细胞移植受者预后算法的深度学习工具

• 批准号: 10286424
• 负责人: Gregory Mark Goldgof
• 金额: $ 11.93万
• 依托单位: UNIVERSITY OF CALIFORNIA, SAN FRANCISCO
• 依托单位国家: 美国
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-08-01 至 2022-08-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10286424
• 关键词: Age; Algorithms; Archives; Artificial Intelligence; Aspirate substance; Autologous Transplantation; Award; Basophilic Erythroblast; Basophils; Biological; Biological Markers; Biology; Biopsy; Blood Cell Count; Blood Cells; Blood Platelets; Bone Marrow; Bone Marrow Cells; Bone Marrow Examination; Bone Marrow Transplantation; Cell Nucleus; Cells; Cellular Morphology; Chromatin; Classification; Clinical; Clinical Management; Clinical Pathology; Clinical Services; Color; Cytoplasmic Granules; Data; Data Set; Descriptor; Detection; Development; Diagnosis; Disease Marker; Engraftment; Erythroblasts; Erythrocytes; Erythroid; Evaluation; Failure; Goals; Hematological Disease; Hematology; Hematopathology; Hematopoiesis; Hematopoietic; Hematopoietic Stem Cell Transplantation; Homologous Transplantation; Human; Human Biology; Image Analysis; Label; Large granular lymphocyte; Lymphoid; Machine Learning; Marrow; Mature Lymphocyte; Measures; Megakaryocytes; Metamyelocyte; Methods; Morphology; Myelocyte; Myelogenous; Normal Cell; Nuclear; Outcome; Pathologist; Pathology; Patient imaging; Patients; Performance; Physiological; Plasma Cells; Polychromatophilic Erythroblast; Population; Process; Progranulocytes; Pronormoblasts; Race; Recovery; Research; Segmented Neutrophil; Services; Shapes; Site; Slide; Specimen; Stains; Standardization; System; Technology; Texture; Time; Training; Transplant Recipients; Transplantation; Vacuole; algorithm training; automated algorithm; automated analysis; base; bone cell; cell type; classification algorithm; classifier algorithm; clinical application; clinical biomarkers; cohort; deep learning; digital; eosinophil; granulocyte; image processing; improved; innovation; insight; learning classifier; learning strategy; monocyte; morphometry; myeloblast; neutrophil; outcome prediction; peripheral blood; post-transplant; prevent; prognostic; sex; success; tool; whole slide imaging

PROJECT SUMMARY/ABSTRACT The bone marrow is the primary site of hematopoiesis and its examination is central to the diagnosis and management of patients with hematological diseases. Pathology slides from clinical hematopathology services represent a treasure trove of real-world data on the biology of human bone marrow. However, human examination is time-intensive and limited in its quantitative precision, preventing our ability to perform large- scale studies or identify new morphologic markers of disease. Machine learning and image processing methods can be applied to the analysis of whole-slide images (WSIs), which will lead to improvements in our understanding of the hematological system, as well as our ability to diagnose and manage hematological diseases. The objective of this research is to build deep learning-based tools for the automated classification of bone marrow aspirates and use these tools to study hematopoiesis in hematopoietic stem cell transplant recipients. The central hypothesis is that automated methods can be developed for the classification, characterization, and quantification of cell morphology on human bone marrow aspirates and that these tools can identify morphologic features predictive of outcome after hematopoietic stem cell transplant. The long-term goal is to develop a suite of artificial intelligence tools for the quantitative analysis of hematopathology WSIs that can be used to improve our understanding of hematopoiesis and our management of patients with hematologic diseases through the development of digital hematopathology tools, stains, and biomarkers for research and clinical applications. This approach is innovative because it applies cutting-edge image analysis technology to the quantitative and scalable study of human bone marrow specimens from clinical pathology archives to drive discovery of new biological insights and clinical biomarkers.

项目摘要/摘要 骨髓是造血的主要部位，其检查是诊断的中心。 以及血液病患者的管理。临床血液病理学中的病理切片 服务代表着有关人类骨髓生物学的真实世界数据的宝库。然而，人类 考试是时间密集型的，其定量精度有限，阻碍了我们进行大规模考试的能力- 对疾病进行规模研究或确定新的形态标志。机器学习与图像处理 这些方法可以应用于全幻灯片图像(WSIS)的分析，这将导致我们的 了解血液系统，以及我们诊断和管理血液系统的能力 疾病。 这项研究的目标是建立基于深度学习的工具，用于自动分类 骨髓抽吸并使用这些工具研究造血干细胞移植中的造血学 收件人。中心假设是可以开发用于分类的自动化方法， 人类骨髓抽提物的细胞形态的表征和量化以及这些工具 可以确定预测造血干细胞移植后结果的形态特征。长期的 目标是开发一套用于血液病理学WSIS定量分析的人工智能工具 这可以用来改善我们对造血的理解和我们对患有 通过开发数字血液病理学工具、染色和生物标记物来治疗血液疾病 研究和临床应用。这种方法是创新的，因为它应用了尖端图像分析 从临床病理角度对人骨髓标本进行定量和可扩展研究的技术 推动发现新的生物学见解和临床生物标记物的档案。

项目成果

Novel computational methods on electronic health record yields new estimates of transfusion-associated circulatory overload in populations enriched with high-risk patients.

电子健康记录的新颖计算方法可以对富含高危患者的人群中与输血相关的循环超负荷进行新的估计。

• DOI: 10.1111/trf.17447
• 发表时间: 2023
• 期刊: Transfusion
• 影响因子: 2.9
• 作者: Wang,Michelle;Goldgof,GregoryM;Patel,Ayan;Whitaker,Barbee;Belov,Artur;Chan,Brian;Phelps,Evan;Rubin,Benjamin;Anderson,Steven;Butte,AtulJ
• 通讯作者: Butte,AtulJ