iScience (Jun 2023)

MRI-based deep learning model predicts distant metastasis and chemotherapy benefit in stage II nasopharyngeal carcinoma

  • Yu-Jun Hu,
  • Lin Zhang,
  • You-Ping Xiao,
  • Tian-Zhu Lu,
  • Qiao-Juan Guo,
  • Shao-Jun Lin,
  • Lan Liu,
  • Yun-Bin Chen,
  • Zi-Lu Huang,
  • Ya Liu,
  • Yong Su,
  • Li-Zhi Liu,
  • Xiao-Chang Gong,
  • Jian-Ji Pan,
  • Jin-Gao Li,
  • Yun-Fei Xia

Journal volume & issue
Vol. 26, no. 6
p. 106932

Abstract

Read online

Summary: Chemotherapy remains controversial for stage II nasopharyngeal carcinoma because of its considerable prognostic heterogeneity. We aimed to develop an MRI-based deep learning model for predicting distant metastasis and assessing chemotherapy efficacy in stage II nasopharyngeal carcinoma. This multicenter retrospective study enrolled 1072 patients from three Chinese centers for training (Center 1, n = 575) and external validation (Centers 2 and 3, n = 497). The deep learning model significantly predicted the risk of distant metastases for stage II nasopharyngeal carcinoma and was validated in the external validation cohort. In addition, the deep learning model outperformed the clinical and radiomics models in terms of predictive performance. Furthermore, the deep learning model facilitates the identification of high-risk patients who could benefit from chemotherapy, providing useful additional information for individualized treatment decisions.

Keywords