Virtual Library

Start Your Search

Sunyi Zheng



Author of

  • +

    P42 - Screening and Early Detection - Risk Modelling and Artificial Intelligence (ID 177)

    • Event: WCLC 2020
    • Type: Posters
    • Track: Screening and Early Detection
    • Presentations: 2
    • Moderators:
    • Coordinates: 1/28/2021, 00:00 - 00:00, ePoster Hall
    • +

      P42.02 - Evaluating the Feasibility of a Deep Learning-Based Computer-Aided Detection System for Lung Nodule Detection in a Lung Cancer Screening Program (ID 739)

      00:00 - 00:00  |  Presenting Author(s): Sunyi Zheng

      • Abstract
      • Slides

      Introduction

      Deep learning techniques have recently achieved remarkable results in automated lung nodule detection. Previously, we developed a deep learning-based computer-aided detection (DL-CAD) system based on the CT scans from the United States. The purpose of this study was to validate its effectiveness for automatic pulmonary nodule detection in an external validation set from a lung cancer screening program.

      Methods

      The proprietary DL-CAD system was pre-trained on the public dataset, LIDC-IDRI, which was collected from seven academic centers resulting in the AI prototype software MIPNOD 1.0. We retrospectively collected 2,127 low-dose CT scans from a Chinese lung cancer screening project between July and December 2017. Three hundred sixty scans were used in this study and they were evaluated independently by radiologists in a double reading fashion and the DL-CAD system. An extra senior radiologist checked all the results and made the consensus as to the reference standard. Free-response Receiver operating characteristic analysis was applied to assess the detection performance of the DL-CAD system.

      Results

      After making the consensus, there were 260 nodules in 196 participants and 164 participants without nodules. The DL-CAD system achieved a sensitivity of 90.0% with one false positive per scan, while radiologists had a sensitivity of 76.5% for detection during double reading. The performance comparison between the DL-CAD system and radiologists in nodule types and Lung-RADS categories was as follows. (1) solid nodules: 90.2% vs 77.2%, P = 0.007; part-solid nodules: 95.2% vs 81.0%, P = 0.810; non-solid nodules: 87.3% vs 72.7%, P = 0.134. (2) Lung-RADS 2: 86.0% vs 66.2%, P < 0.001; Lung-RADS 3: 94.3% vs 90.0%, P = 0.549; Lung-RADS 4: 100.0% vs 97.0%, P = 1.000.

      Conclusion

      The deep learning-based CAD system showed good performance for automatic pulmonary nodule detection on an external dataset and could provide assistance for radiologists in lung cancer screening programs.

      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.

    • +

      P42.06 - Automatic Lung Nodule Detection by a Deep Learning-Based CAD System: The Value of Slab Thickness in the Maximum Intensity Projection Technique (ID 737)

      00:00 - 00:00  |  Presenting Author(s): Sunyi Zheng

      • Abstract
      • Slides

      Introduction

      Deep learning techniques have shown good results on automated lung nodule detection. The maximum intensity projection technique (MIP) has been proved that it can improve the performance of nodule detection in the deep learning-based computer-aided detection (DL-CAD) system. But the optimal slab thickness is unknown. To provide better assistance for radiologists, the aim of this study is to investigate the effect of the slab thickness in maximum intensity projections by a DL-CAD system on pulmonary nodule detection in CT scans.

      Methods

      The public LIDC-IDRI dataset includes 888 CT scans with 1186 nodules annotated by four radiologists. From those scans, MIP images were reconstructed with slab thicknesses of 5 to 50 mm (at 5 mm intervals) and 3 to 13 mm (at 2 mm intervals). The DL-CAD system was trained separately using MIP images with various slab thicknesses. Based on ten-fold cross-validation, the sensitivity and the F2 score were determined to evaluate the performance of the DL-CAD system for nodule detection.

      Results

      The combination of results from 16 MIP slab thickness settings showed a high sensitivity of 98.0%. The sensitivity increased (82.8% to 90.0%) for slab thickness of 1 to 10 mm and decreased (88.7% to 76.6%) for slab thickness of 15 to 50 mm. The number of false positives (FPs) was decreasing with increasing slab thickness, but was stable at 4 FP/scan at a slab thickness of 30 mm or more. With a MIP slab thickness of 10 mm, the DL-CAD system reached the highest sensitivity of 90.0%, with 8 FPs/scan.

      Conclusion

      The utilization of multi-MIP images could further improve the performance of lung nodule detection in the DL-CAD system. The DL-CAD system showed the highest sensitivity for pulmonary nodule detection based on MIP images of 10 mm, similar to the slab thickness usually applied by radiologists.

      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.