Virtual Library

Start Your Search

Narges Razavian

Author of

  • +

    P1.09 - Pathology (Not CME Accredited Session) (ID 941)

    • Event: WCLC 2018
    • Type: Poster Viewing in the Exhibit Hall
    • Track:
    • Presentations: 1
    • Moderators:
    • Coordinates: 9/24/2018, 16:45 - 18:00, Exhibit Hall
    • +

      P1.09-32 - Classification and Mutation Prediction from Non-Small Cell Lung Cancer Histopathology Images Using Deep Learning (ID 12631)

      16:45 - 18:00  |  Author(s): Narges Razavian

      • Abstract


      Visual inspection of histopathology slides of lung tissues is one of the primary methods used by pathologists to assess pathological stage and classifications of lung cancer. Adenocarcinoma and squamous cell carcinoma are the two most prevalent types of non-small cell lung cancer (NSCLC), but their distinction can be challenging and time-consuming even for the expert eye. Furthermore, the increasing prevalence of new treatment modalities relies on classification of NSCLC mutational analyses of the tumor to guide targeted therapy and provide prognostic information.

      a9ded1e5ce5d75814730bb4caaf49419 Method

      In this study, we trained a deep convolutional neural network (CNN) model (inception v3) on histology images obtained from The Cancer Genome Atlas (TCGA) to accurately and automatically classify whole-slide images into adenocarcinoma, squamous cell carcinoma or normal lung tissue. Furthermore, we trained the neural network to predict the ten most commonly mutated genes in lung adenocarcinoma. Our model was then validated on independent datasets of images obtained from both frozen and formalin-fixed paraffin-embedded tissue, including resection and biopsy specimens.

      4c3880bb027f159e801041b1021e88e8 Result

      We found that the performance of our method is comparable, in terms of sensitivity and specificity, to that of pathologists, with a 0.97 average Area Under the Curve (AUC) on a held-out population of whole-slide images for the sub-type classification on frozen sections. Additionally, we found that six of the ten most commonly mutated genes – STK11, EGFR, FAT1, SETBP1, KRAS and TP53 – can be predicted by the algorithm solely using information from histologic images with an accuracy ranging from 0.733 to 0.856, as measured by the AUC on the held-out population.

      8eea62084ca7e541d918e823422bd82e Conclusion

      These findings suggest that deep learning models can offer both pathologists and patients a fast, accurate, and inexpensive classification of NSCLC or gene mutations, and thus have a significant impact on cancer treatment. Our computational approach can be applied to any cancer type and the code is available as free open-source software at