Convergent AI for Precise Breast Cancer Risk Assessment

摘要 乳腺癌仍然是美国妇女癌症死亡的主要原因之一。 尽管在乳腺癌预后和预测标志物的鉴定方面取得了进展， 治疗乳腺X线检查报告是筛查和诊断乳腺癌的第一步。异常 乳房X线检查结果，如肿块、异常钙化、结构扭曲和不对称 密度可以导致癌症诊断。美国放射学会开发了乳腺成像 报告和数据系统（BI-RADS）词典，用于标准化乳腺X线摄影报告，以便于活检 决策的然而，BI-RADS词汇的应用导致了大量的观察者间的差异。 变异性，包括术语使用不当和数据缺失。这种观察者的可变性部分导致了 美国各地的活检率存在相当大的差异，大多数乳腺活检最终发现， 良性病变。因此，需要一种系统，可以更好地对癌症风险进行分层，并定义一个 活检的最佳阈值。为了满足这一需求，我们建议开发一种智能增强的风险 基于多模态图像和临床信息的乳腺癌管理评估系统 深度学习和数据挖掘技术。 本研究旨在开发一个定义明确、新颖的风险评估系统， 具有新型预测模型的数据集，该模型输出的癌症概率测量值在临床上更准确， 相关性和信息量比六个离散的BI-RADS评分。使用乳房X线摄影或乳房超声BI- RADS报告签名和放射组学特征，一种更精确和临床上更有效的预测模型 相关性可能被开发为针对充分表征和定义的特定活检患者亚组，而不是 而不是广泛的异质性活检组。我们提出的技术需要一种新的策略， 语言处理从大量的数据中提取与乳腺癌相关的临床风险因素。 患者图表自动化，并将其与相应的图像组学数据和放射科医生整合在一起， 生成的报告。我们将从大量的乳房X线摄影中提取和量化图像特征， 和乳房超声图像，并将其与放射学报告和相关的临床风险概况相结合 和其他患者特征来生成风险评估评分，以帮助放射科医生和肿瘤科医生 乳腺癌风险评估和活检决策。这样一个基于网络的应用工具将是第一个乳房 癌症风险评估系统基于AI方法增强的综合放射组学数据。iBRISK 该工具将增强患者和临床医生之间的互动，以便就是否或 而不是活检 我们的假设是BI-RADS报告和成像指标包含乳腺的重要特征 癌症风险评估和活检决策。通过使用BI-RADS报告和成像指标，我们 将能够开发新的指标，以更好地评估乳腺癌风险。乳腺癌的新奇 风险评估系统将采用一种新的预测模型，部署深度学习和人工智能， 技术，为乳腺癌风险评估提供更可靠的BI-RADS亚型分层 减少不必要的乳腺活检和患者的焦虑。

ABSTRACT Breast cancer continues to be one of the leading causes of cancer death among women in the United States, despite the advances made in the identification of prognostic and predictive markers for breast cancer treatment. Mammographic reporting is the first step in the screening and diagnosis of breast cancer. Abnormal mammographic findings such as a mass, abnormal calcifications, architectural distortion, and asymmetric density can lead to a cancer diagnosis. The American College of Radiology developed the Breast Imaging Reporting and Data System (BI-RADS) lexicon to standardize mammographic reporting to facilitate biopsy decision-making. However, application of the BI-RADS lexicon has resulted in substantial inter-observer variability, including inappropriate term usage and missing data. This observer variability has lead in part to a considerable variation in the rate of biopsy across the US, with a majority of breast biopsies ultimately found to be benign lesions. Hence, there is the need for a system that can better stratify the risk of cancer and define a more optimum threshold for biopsy. To address this need, we propose to develop an intelligent-augmented risk assessment system for breast cancer management based on multimodality image and clinical information with deep learning and data mining techniques. This study aims to develop a well-defined, novel risk assessment system incorporating multi-modality datasets with a novel predictive model that outputs a probability measure of cancer that is more clinically relevant and informative than the six discrete BI-RADS scores. Using mammographic or breast ultrasound BI- RADS reporting signatures and radiomics features, a predictive model that is more precise and clinically relevant may be developed to target well-characterized and defined specific biopsy patient subgroups rather than a broad heterogeneous biopsy group. Our proposed technique entails a novel strategy using Natural Language Processing to extract pertinent clinical risk factors related to breast cancer from vast amounts of patient charts automatically and integrate them with corresponding image-omics data and radiologist- generated reports. We will extract and quantitate image features from both large amounts of mammography and breast ultrasound images and combine them with the radiology reports and pertinent clinical risk profile and other patient characteristics to generate a risk assessment score to aid radiologists and oncologists in breast cancer risk assessment and biopsy decisions. Such a web-based application tool will be the first breast cancer risk assessment system based on integrative radiomics data augmented by AI methods. The iBRISK tool will enhance engagement between the patient and clinician for making an informed decision on whether or not to biopsy. Our hypothesis is that BI-RADS reports and the imaging metrics contain significant features for the breast cancer risk assessment and biopsy decision-making. By using BI-RADS reports and the imaging metrics, we will be able to develop new metrics to better breast cancer risk assessment. The novelty of the breast cancer risk assessment system is that it will incorporate a new predictive model that deploys deep learning and AI technology to provide a more reliable stratification of the BI-RADS subtypes for breast cancer risk assessment and reduce unnecessary breast biopsies and patients’ anxiety.