Deep learning-based prognosis prediction in cancer patients

2020 Fiscal Year Final Research Report

• Project/Area Number: 19K17208
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 52040:Radiological sciences-related
• Research Institution: Juntendo University
• Principal Investigator: Inoue Tatsuya 順天堂大学, 医学部, 助教 (00733362)
• Project Period (FY): 2019-04-01 – 2021-03-31
• Keywords: 深層学習 / Radiomics / 放射線治療 / 人工知能 / 予後予測

Outline of Final Research Achievements

Accurate prediction of patient-specific response to the cancer treatment is important for personalized medicine. Imaging-based prediction methods can generally be categorized into 2 types: handcrafted-feature-based radiomics methods and automatically self-learned-feature-based deep learning methods, which has achieved state-of-the-art performances on image recognition recently.
In this work, we generated various types of deep learning model such as multi-input model, multi-task model and multimodal model for the prediction of prognosis in patients with lung cancer or head and neck cancer, and investigated the performances.

Free Research Field

医学物理学

Academic Significance and Societal Importance of the Research Achievements

現在の癌治療は癌種や病期などに基づいて画一的に治療法が決定されている。しかし、放射線や抗がん剤を使った癌治療の効果は、例え同じ治療でも個々人が持つ背景に影響を受け、ある患者に有効であった治療法が別の患者にも効果的であるとは限らない。
治療を実施する前に取得する画像データから治療の効果を予測出来ることは、患者毎に治療効果を最大化する治療方法の決定や開発に繋がることが期待される。