A Deep Learning Model to Improve Pathologist Interpretation of Donor Kidney Biopsies

改善病理学家对供体肾活检的解释的深度学习模型

• 批准号: 10266188
• 负责人: Joseph P Gaut
• 金额: $ 77.51万
• 依托单位: NEWVENTUREIQ, LLC
• 依托单位国家: 美国
• 财政年份: 2018
• 资助国家: 美国
• 起止时间: 2018-09-21 至 2023-08-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10266188
• 关键词: Adoption; Americas; Artificial Intelligence; Biopsy; Canada; Cessation of life; Chronic; Cicatrix; Clinical; Computer software; Computers; Contracts; Data; Databases; Development; Ensure; Evaluation; Fast Healthcare Interoperability Resources; Fibrosis; Frozen Sections; Funding; Goals; Gold; Graft Survival; Health Care Costs; Human; Interobserver Variability; Kidney; Kidney Diseases; Kidney Transplantation; Knowledge; Laboratories; Letters; Life; Machine Learning; Malignant neoplasm of prostate; Manuals; Measurement; Microscope; Microscopic; Midwestern United States; Modeling; Multivariate Analysis; Online Systems; Organ; Organ Donor; Organ Procurements; Outcome; Pathologist; Pathology; Patient-Focused Outcomes; Patients; Performance; Personal Satisfaction; Phase; Process; Reproducibility of Results; Research Personnel; Savings; Scanning; Scientist; Secure; Services; Slide; Small Business Technology Transfer Research; Specialist; Speed; System; Techniques; Testing; Tissues; Transplantation; Trichrome stain; Trichrome stain method; Trust; United Network for Organ Sharing; Universities; Variant; Washington; Work; analytical tool; base; clinical biomarkers; cloud based; cloud platform; commercial application; cost; deep learning; functional improvement; glomerulosclerosis; image processing; imaging biomarker; improved; innovation; interstitial; kidney biopsy; learning strategy; malignant breast neoplasm; pathology imaging; phase 1 study; predictive modeling; public health relevance; renal damage; shared database; standard of care; technological innovation; tool; whole slide imaging

ABSTRACT More people die every year from kidney disease than breast or prostate cancer. Kidney transplantation is life-saving, yet the donor organ shortage and high organ discard rate contributes to 13 deaths daily among patients awaiting transplant. The decision to use or discard a donor kidney relies heavily on microscopic quantitation of chronic damage by pathologists. The current standard of care relies on a manual process that is subject to significant human variability and inefficiency, resulting in potentially healthy kidneys being discarded and potentially damaged kidneys being transplanted inappropriately. Our team developed the first Deep Learning model to quantify percent global glomerulosclerosis in donor kidney frozen section biopsy whole slide images. We developed a cloud-based platform to apply the Deep Learning model to analyze kidney biopsy whole slide images in under 6 minutes with accuracy and precision equal to or greater than current standard of care pathologists. We have also developed a Deep Learning model to quantify interstitial fibrosis on donor kidney biopsy whole slide images. This innovative approach has the potential to transform donor kidney biopsy evaluation by improving pathologist efficiency, accuracy, and precision ultimately resulting in optimized donor organ utilization, improved patient outcomes, and diminished health care costs. The goal of this project is to establish our Deep Learning automated techniques as the standard for evaluating donor kidneys prior to transplantation. This will be achieved by assembling a team of expert pathologists and computer scientists specializing in machine learning. The proposal will evaluate the accuracy and precision of the interstitial fibrosis Deep Learning model, use the automated quantitation of key microscopic findings to develop an outcome-based chronic damage score that predicts graft outcome, and test the ability of the Deep Learning models to withstand variations encountered using different scanners and processing in different laboratories. The functionality of the Trusted Kidney software platform will be improved beyond the current usable product into a commercially viable solution for multiple laboratories.

抽象的 每年死于肾病的人数多于乳腺癌或前列腺癌。肾 移植挽救生命，但供体器官短缺、器官废弃率高 每天有 13 名等待移植的患者死亡。决定使用或 丢弃供体肾脏在很大程度上依赖于慢性损伤的显微定量 病理学家。当前的护理标准依赖于手动过程，该过程受 显着的人类变异性和低效率，导致潜在的健康肾脏 被丢弃和可能受损的肾脏被不当移植。我们的团队 开发了第一个深度学习模型来量化捐赠者的整体肾小球硬化百分比 肾脏冰冻切片活检全幻灯片图像。我们开发了一个基于云的平台来应用 深度学习模型可在 6 分钟内分析肾活检整个幻灯片图像 准确性和精密度等于或高于病理学家当前的护理标准。我们有 还开发了深度学习模型来量化供体肾活检的间质纤维化 整个幻灯片图像。这种创新方法有可能改变供体肾活检 通过提高病理学家的效率、准确性和精密度进行评估，最终导致 优化供体器官的利用，改善患者的治疗效果，并降低医疗保健成本。 该项目的目标是将我们的深度学习自动化技术建立为标准 用于在移植前评估供体肾脏。这将通过组建一个团队来实现 由专门从事机器学习的病理学家和计算机科学家组成。提案 将评估间质纤维化深度学习模型的准确性和精确度，使用 对关键的微观发现进行自动定量，以开发基于结果的慢性病 预测移植结果的损伤评分，并测试深度学习模型的能力 承受使用不同扫描仪和不同处理方式时遇到的变化 实验室。 Trusted Kidney软件平台的功能将得到改进 将当前可用的产品转变为多个实验室商业上可行的解决方案。