2023 Fiscal Year Final Research Report
Prediction of prognosis after radiochemotherapy for head and neck cancer using recurrent neural networks and MR images
Project/Area Number |
21K15814
|
Research Category |
Grant-in-Aid for Early-Career Scientists
|
Allocation Type | Multi-year Fund |
Review Section |
Basic Section 52040:Radiological sciences-related
|
Research Institution | St. Marianna University School of Medicine |
Principal Investigator |
Tomita Hayato 聖マリアンナ医科大学, 医学部, 講師 (90647801)
|
Project Period (FY) |
2021-04-01 – 2024-03-31
|
Keywords | 深層学習 / 人工知能 / 予後予測 / 喉頭癌 / 下咽頭癌 / 拡散強調画像 / 放射線治療 |
Outline of Final Research Achievements |
This preliminary study aimed to develop a DL model using DWI and ADC map to predict local recurrence and 2-year PFS in laryngeal and hypopharyngeal cancer patients treated by curative therapy related to radiotherapy. All patients underwent MR before and 4 weeks after the start of radiotherapy. The DL models that extracted imaging features on pretreatment and intra-treatment DWI and ADC map were trained to predict the local recurrence within a 2-year follow-up. The best AUC and accuracy for predicting the local recurrence in the DL model using intra-treatment DWI (DWIintra) were 0.767 and 81.0 %, respectively. Log-rank test showed that DWIintra was significantly associated with PFS (P = 0.013). DWIintra was an independent prognostic factor for PFS in multivariate analysis (P = 0.016).
|
Free Research Field |
放射線診断
|
Academic Significance and Societal Importance of the Research Achievements |
畳み込み型ニューラルネットワークと治療中の拡散強調像を用いた場合、治療後の再発はAUCで0.767,正診率は81.0%であった。高リスク群と低リスク群に分類し、 Log-rankテストを行うと、同手法を用いた治療中の拡散強調像から2年の予後予測であった(P = 0.013)。また、Cox regression解析では深層学習による手法が 2年の予後予測の唯一の因子であることが分かった(P = 0.016)。このことは、これまで医用画像と深層学習を用いた研究では証明されていなかった内容である。また、本研究は他の腫瘍でも同様の方法を使用することができるため、研究の意義は大きいと思われる。
|