RSI-AI: Predicting clinically significant prostate cancer to guide biopsy decisions by combining advanced tissue microstructure imaging with deep learning

RSI-AI：通过将先进的组织微观结构成像与深度学习相结合来预测具有临床意义的前列腺癌，以指导活检决策

• 批准号: 10254808
• 负责人: Nathan Scott White
• 金额: $ 25.17万
• 依托单位: CORTECHS LABS, INC.
• 依托单位国家: 美国
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-08-15 至 2023-07-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10254808
• 关键词: 3-Dimensional; Affect; Age; Algorithms; Biopsy; Blinded; Classification; Clinical; Clinical Data; Clinical Treatment; Communities; Computer software; Data; Data Set; Detection; Diagnosis; Diagnostic; Diffusion Magnetic Resonance Imaging; Early Diagnosis; Family; Future; Genetic; Gland; Gleason Grade for Prostate Cancer; Goals; Health Care Costs; Image; Imaging Techniques; Indolent; Label; Legal patent; Lesion; Location; Magnetic Resonance Imaging; Malignant - descriptor; Malignant Neoplasms; Malignant neoplasm of prostate; Medical Device; Methods; Modeling; Output; PSA screening; Pathologic; Pathology; Performance; Phase; Production; Prostate; Quality of life; Radical Prostatectomy; Radiology Specialty; Reader; Recording of previous events; Restriction Spectrum Imaging; Risk Assessment; Risk Factors; Scanning; Screening for Prostate Cancer; Sensitivity and Specificity; System; Testing; Tissues; Training; United States; Update; Validation; accurate diagnosis; base; cancer site; clinical predictors; clinical risk; clinically significant; cloud platform; commercialization; convolutional neural network; deep learning; demographics; imaging modality; improved; innovation; men; neural network; neural network architecture; noninvasive diagnosis; novel; overtreatment; prospective; prostate biopsy; prostate cancer risk; prostate lesions; radiologist; risk prediction model; safety net; software development; spectrograph; tool; tumor; usability; virtual biopsy; wasting; water diffusion

Prostate biopsies are critical for the diagnosis of prostate cancer, but it is often unclear who should be biopsied and where in the gland the biopsy should be targeted. This results in missed diagnoses, unnecessary biopsies, and overdiagnosis and overtreatment of cancer that is not life threatening. The goal of this proposal is to develop a set of quantitative and non-invasive tools, RSI-AI and RSI-AI+, to help clinicians determine who should be biopsied for prostate cancer and the locations of clinically significant lesions. RSI-AI uses deep learning to predict the location and pathological grade of prostate cancer lesions from restriction spectrum imaging (RSI) data. RSI is an advanced diffusion magnetic resonance imaging (MRI) technique that models the restricted diffusion of water molecules to improve microtissue classification and tumor detection. By utilizing RSI data in the deep learning model, RSI-AI will produce pathological grade predictions that are more accurate than models trained with conventional MRI data. RSI-AI+ integrates the pathological grade predictions from RSI-AI with clinical data including age, family history, genetics, and prostate volume to accurately and comprehensively quantify current and future risk for prostate cancer. Phase I of this proposal will develop and validate the RSI-AI and RSI-AI+ models and compare their performance to models trained with conventional MRI data. Phase II of this proposal will deploy RSI-AI and RSI-AI+ to the Cortechs cloud platform, demonstrate their clinical usability and utility, and generate the materials required for a 510K FDA submission. The clinical software generated through this proposal will ultimately improve diagnostic yields, reduce unnecessary biopsies and overtreatment of indolent prostate cancer, while facilitating early detection and appropriate treatment of clinically significant prostate cancer

前列腺活检对于前列腺癌的诊断至关重要，但通常不清楚是谁 应该进行活检，活检应该针对腺体的哪个部位。这导致错过 诊断，不必要的活检，过度诊断和过度治疗癌症，而不是生命 威胁。该提案的目标是开发一套定量和非侵入性工具， RSI-AI和RSI-AI+，帮助临床医生确定哪些人应该接受前列腺癌活检 以及临床上显著病变的位置。RSI-AI使用深度学习来预测 限制性频谱成像对前列腺癌病变的定位和病理分级 (RSI)数据RSI是一种先进的弥散磁共振成像（MRI）技术， 对水分子的受限扩散进行建模，以改善微组织分类， 肿瘤检测通过在深度学习模型中利用RSI数据，RSI-AI将产生 病理分级预测比用常规MRI训练的模型更准确 数据RSI-AI+将RSI-AI的病理分级预测与临床数据相结合 包括年龄、家族史、遗传学和前列腺体积， 量化前列腺癌的当前和未来风险。该提案的第一阶段将制定和 验证RSI-AI和RSI-AI+模型，并将其性能与使用 常规MRI数据。本提案的第二阶段将在 Cortechs云平台，展示其临床可用性和实用性，并生成 510 K FDA申请所需的材料。通过此生成的临床软件 该提案将最终提高诊断率，减少不必要的活检， 过度治疗惰性前列腺癌，同时促进早期发现和适当的 临床上显著的前列腺癌的治疗