Virtual Library

Start Your Search

Junfeng Xiong



Author of

  • +

    P02 - Diagnostics and Interventional Pulmonology (ID 110)

    • Event: WCLC 2020
    • Type: Posters
    • Track: Diagnostics and Interventional Pulmonology
    • Presentations: 1
    • Moderators:
    • Coordinates: 1/28/2021, 00:00 - 00:00, ePoster Hall
    • +

      P02.01 - Evaluation of PFS for Pulmonary Adenocarcinoma Patient Treated With EGFR Tyrosine Kinase Inhibitors Using Deep Learning Based on CT Image (ID 987)

      00:00 - 00:00  |  Presenting Author(s): Junfeng Xiong

      • Abstract
      • Slides

      Introduction

      For stage IV patients harboring epidermal growth factor receptor (EGFR) mutations, tyrosine kinase inhibitor (TKI) is recommended to be the first-line treatment modality especially. But all patients harboring EGFR mutations respond differently to first-line TKI treatment and PFS might range from 2 to 15 months, consequently. In this study, we aimed to establish a deep learning model to predict the PFS in pulmonary adenocarcinoma patients with harboring EGFR mutations and receiving TKI.

      Methods

      349 patients, who were diagnosed with stage IIIB&IV lung adenocarcinomas harboring EGFR mutations and received first-line TKI treatment, were collected from 2013 to 2017. Tumors with a ground-glass component or diameter less than 8 mm were excluded. The CT images were taken before receiving treatment. Region of interest that included tumors were segmented manually by radiologists. Patients received completed follow-up every three months to confirm the exact PFS. Patients with favorable PFS (>9 months) were regarded as positive samples, while the others were negative samples. Patients were randomly divided into training set (n=249) and validation set (n=100). A deep learning model based on 3D CNN was built by using the training set and transfer learning method. Incorporated with clinical features, including gender, age, smoking history, TNM stage, and EGFR mutation point, the performance of the deep learning model was further improved. Model performance was evaluated on the validation dataset using the area under receiver operating characteristic curve (AUC) and Kaplan-Meier survival analysis.

      Results

      The baseline prediction model based on clinical features achieved AUC of 0.624 (95% CI: 0.510, 0.739). The deep learning models without or with clinical features showed AUCs of 0.701 (95% CI: 0.575-0.785) and 0.757 (95% CI: 0.631, 0.816), respectively. Kaplan-Meier survival analysis showed that our model can significantly (log-rank test, P<0.001) distinguish the difference in PFS between the high- and low-risk subgroups divided by the prediction results of our 3D CNN model.

      figure 1.jpg

      Conclusion

      The proposed deep learning model incorporated with clinical features has a potential ability to predict the PFS in lung adenocarcinomas patients receiving first-line TKI treatment and may help to make clinical decisions.

      Only Active Members that have purchased this event or have registered via an access code will be able to view this content. To view this presentation, please login or select "Add to Cart" and proceed to checkout.