Virtual Library

Start Your Search

Junfeng Xiong



Author of

  • +

    MA 14 - Diagnostic Radiology, Staging and Screening for Lung Cancer I (ID 672)

    • Event: WCLC 2017
    • Type: Mini Oral
    • Track: Radiology/Staging/Screening
    • Presentations: 1
    • +

      MA 14.12 - Detecting Epidermal Growth Factor Receptor Mutation Status in Patients with Lung Adenocarcinoma Using Radiomics and Random Forest (ID 9772)

      15:45 - 17:30  |  Author(s): Junfeng Xiong

      • Abstract
      • Presentation
      • Slides

      Background:
      We tried a radiomics approach to build a random forest classifier for recognition of epidermal growth factor receptor (EGFR) mutation status in Chinese patients with lung adenocarcinomas using quantitative image features extracted from non-enhanced computed tomography (CT) images

      Method:
      From October 2008 to December 2015, 355 patients diagnosed with lung adenocarcinomas were included in this retrospective study. They all have complete clinical, pathological, and EGFR mutation status information, and their CT images were scanned before any invasive operation. Tumors with ground glass component or diameter smaller than 2 cm were not included. Their pathological phenotypes and EGFR mutation status were gained from surgical resections. Region of tumors on CT images were segmented semi-automatically first then manually modified by experienced clinicians. 440 quantitative image features were extracted from CT images and fall into four groups: first order statistics, shape and size based features, textural features, and wavelet features. Random forest was used to build the classification model which takes all the features into consideration and make an overall probability of mutation based on the vote of decision trees. The random forest classifier was validated using an independent set and its performance was evaluated using area under curve (AUC) values of the receiver operating characteristic

      Result:
      355 patients diagnosed with lung adenocarcinoma were enrolled in this study (170 male, 185 female; 54 smokers, 301 non-smokers). The patients all received surgery based treatment and their tumor stage varied from I to IV. EGFR mutations (mainly 19del and 21L858R) were found in 187/285(65.6%) and 48/70(68.6%) patients in training and validation sets respectively. The random forest model showed an AUC of 0.781 (95% confidence interval: 0.668-0.894, p<0.001) in the validation set. The sensitivity and specificity are 60.4% and 90.9% at best diagnostic decision point. These results were highest among published results of only using images to detect EGFR.

      Conclusion:
      The random forest classifier based on CT images showed potential ability to identify EGFR mutations in patients with lung adenocarcinomas and could be improved in future works.

      Only 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, select "Add to Cart" and proceed to checkout. If you would like to become a member of IASLC, please click here.

      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.

  • +

    P3.13 - Radiology/Staging/Screening (ID 729)

    • Event: WCLC 2017
    • Type: Poster Session with Presenters Present
    • Track: Radiology/Staging/Screening
    • Presentations: 1
    • +

      P3.13-022 - 3D CNNs for Recognition of Epidermal Growth Factor Receptor Mutation Status in Patients with Lung Adenocarcinoma (ID 9799)

      09:30 - 16:00  |  Presenting Author(s): Junfeng Xiong

      • Abstract
      • Slides

      Background:
      In this study, we built three 3-dimensional convolutional neural networks (CNN) for recognition of epidermal growth factor receptor (EGFR) mutation status in Chinese patients with lung adenocarcinomas based on non-enhanced computed tomography (CT) images.

      Method:
      From October 2008 to December 2015, 405 patients with lung adenocarcinomas were included in this retrospective study. Their pathological phenotypes and EGFR mutation status were gained from surgical resections. Their CT images used in this study were taken before any invasive operation. Tumors with a diameter smaller than 8 mm or have ground glass component were excluded. Region of interest that includes tumors were segmented manually by clinicians and preprocessed to have uniform size and grey-level range before applied to CNNs. The three CNNs have 4 convolutional and 1 full connection layers between input and output layers. The inputs size of three CNNs are 21×21×21, 31×31×31, and 41×41×41, respectively. The outputs of the CNN are the probabilities of mutant and wild status. The CNN classifier’s performance was then validated using an independent set and evaluated using area under curve (AUC) values of the receiver operating characteristic.

      Result:
      405 patients diagnosed with lung adenocarcinoma staging I to IV were included in this study (195 male, 210 female; 61 smokers, 344 non-smokers). The patients received surgery based treatment and their tumor stage was based on pathological reports. EGFR mutations (mainly 19del and 21L858R) were found in 198/320(61.9%) and 56/85(65.9%) patients in training and validation sets, respectively. The CNN showed an AUC of 0.767 (95% confidence interval: 0.668-0.866, p<0.001) in the validation set. The sensitivity and specificity are 62.5% and 89.7% at best diagnostic decision point. These results were highest among published results of only using images to recognize EGFR.

      Conclusion:
      The CNN showed potential ability to recognize EGFR mutation status in patients with lung adenocarcinomas and could be improved in the future works to help 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.