Development and Clinical Implementation of an Artificial Intelligence Tool to Predict Risk of Upgrade of Ductal Carcinoma In Situ

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ABSTRACT This proposal presents a five-year career development plan focused on data science and artificial intelligence (AI) and the application of AI to improve outcomes in women with ductal carcinoma in situ (DCIS). The candidate is a Radiologist at MGH and an Assistant Professor of Radiology at Harvard Medical School. The proposal builds upon the candidate’s previous research and clinical experiences in breast imaging and also upon a strong ongoing research partnership between MGH and MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL). The candidate’s long-term career goal is to become a leader in academic breast imaging by investigating and applying AI to critical areas in breast cancer detection, diagnosis, and treatment. The proposed research project and advanced didactic training at Harvard and MIT will position the candidate with a unique set of knowledge and skills in data science and AI that will enable her to develop an independent cancer research program that focuses on applications of AI to breast imaging. The incidence of DCIS has dramatically increased over the past 40 years, with an estimated 63,960 diagnoses in 2018. Current guidelines recommend that DCIS be treated with surgery, radiation, and endocrine therapy, but there remains considerable controversy over whether this regimen represents overtreatment for those women with indolent non-hazardous DCIS. Given concerns about overtreatment, there are currently three randomized controlled trials underway to evaluate the safety and efficacy of active surveillance versus standard treatment, and critical to the implementation of active surveillance programs is careful selection of eligible patients. The goal of the proposed project is to develop a robust AI tool that incorporates clinical data, mammographic imaging, and biopsy histopathology slides for pre-operatively predicting the risk of concurrent invasive cancer in women with DCIS. The tool will be built using machine learning, deep learning, and computer vision. Incorporation of mammographic imaging and histopathology slides into the AI tool will be supported by the MGH & BWH Center for Clinical Data Science (CCDS) and the MGH Department of Pathology. After development and validation of the AI tool based on a retrospective cohort of 1,400 women diagnosed with DCIS at MGH, the tool will then be integrated into MGH’s mammography information system and used to categorize new cases of DCIS. The specific aims are: (1) to develop a robust AI tool that predicts the risk of upgrade of DCIS diagnosed by image-guided core needle biopsy to invasive cancer at surgery and (2) to implement and evaluate the AI tool in clinical practice. Use of this tool could identify the subset of women who are appropriate candidates for active surveillance, decrease the morbidity and costs of overtreatment, and support more targeted and precise treatment options for women diagnosed with DCIS.