Author of

• 11645

  +

  P2.08 - Poster Session 2 - Radiotherapy (ID 198)

  • Event: WCLC 2013
  • Type: Poster Session
  • Track: Radiation Oncology + Radiotherapy
  • Presentations: 1
  • Moderators:
  • Coordinates: 10/29/2013, 09:30 - 16:30, Exhibit Hall, Ground Level

  • 11672

    +

    P2.08-027 - Atlas segmentation for lung stereotactic ablative radiotherapy (ID 3195)

    09:30 - 09:30  |  Author(s): J. Sykes
    • Abstract

    Background
    With the high radiation dose used in stereotactic ablative body radiotherapy (SABR), it is vital to accurately delineate radiosensitive normal tissues in the planning process such that the dose distributions can be optimised to avoid these regions. This can become a time consuming process. It is also well understood that there is variation in how these volumes are delineated and this has influence on the plan and follow-up data. This work aims to determine if the use of automatic atlas segmentation can help to quickly and accurately delineate CT volumes for SABR planning - saving time, reducing complexity and improving the uniformity of process between patients.

    Methods
    From a retrospective sample of 5 patients delineated by an oncologist, each patient’s CT scan and contours was used as an atlas for a deformable image registration to map volumes to the other patients. A range of quantitative metrics (based on both overlap and distance) were used to evaluate the agreement between the expert contours and the atlas contours for each.

    Results
    The Dice similarity co-efficient showed overlap varied depending on the structure – lungs and heart performed well with a mean of 90% (range 71 – 99%) while the airway structures (trachea, oesophagus, proximal bronchi) performed poorer, with a mean of 52% (range 17 – 99%). This indicates while some structures are well suited to atlas segmentation, some are not. The mean Hausdorff distance, indicated that contours requiring editing after applying the atlas, had mean Hausdorff distances on the order of 1 mm. Typically, atlases that did not perform acceptably had a mean Hausdorff distance > 3 mm indicating it was better to start again with manual delineation. From the sample of 5 atlases, none performed significantly better than the others across all cases.

    Conclusion
    The use of atlases could potentially encourage consistency in delineation and could reduce the laborious task of delineation for lung SBRT treatment planning, however editing of contours is still required. Larger structures performed well, however they could also be easily delineated with threshold-based automatic segmentation. Further investigation to develop a bank of representative atlases may improve performance.